Skip to contents

Continuous summary table

Source: R/table_continuous.R
table_continuous.Rd

Computes descriptive statistics (mean, SD, min, max, confidence interval of the mean, n) for one or many continuous variables selected with tidyselect syntax.

With by, produces grouped summaries and reports a group-comparison p-value by default (Welch test; change via test). Additional inferential output is opt-in: test statistics (statistic) and effect sizes (effect_size / effect_size_ci). Set p_value = FALSE to suppress the p-value column. Without by, produces one-way descriptive summaries.

Multiple output formats are available via output: a printed ASCII table ("default"), a plain data.frame ("data.frame", the displayed table with its numbers, or "long", every statistic of each variable and group; see Value), or publication-ready tables ("tinytable", "gt", "flextable", "excel", "clipboard", "word").

This is the descriptive companion to table_continuous_lm(). The two functions share their layout, alignment, and reporting precision so descriptive and model-based analyses of the same data look uniform side by side – with one exception, documented under align: only table_continuous() carries align into the excel output. Use table_continuous_lm() when you need robust SE, weighted contrasts, fitted means, or covariate adjustment.

Usage

table_continuous(
  data,
  select = tidyselect::everything(),
  by = NULL,
  exclude = NULL,
  regex = FALSE,
  drop_na = TRUE,
  weights = NULL,
  rescale = getOption("spicy.rescale", FALSE),
  test = c("welch", "student", "nonparametric"),
  p_value = NULL,
  p_adjust = "none",
  statistic = FALSE,
  show_n = TRUE,
  show_columns = NULL,
  effect_size = c("none", "auto", "hedges_g", "eta2", "r_rb", "epsilon2"),
  effect_size_ci = FALSE,
  smd = FALSE,
  ci = TRUE,
  labels = getOption("spicy.var_labels", FALSE),
  note_labels = NULL,
  ci_level = 0.95,
  digits = 2,
  effect_size_digits = 2,
  p_digits = 3,
  decimal_mark = ".",
  align = c("decimal", "center", "right"),
  note = NULL,
  note_add = NULL,
  output = c("default", "data.frame", "long", "tinytable", "gt", "flextable", "excel",
    "clipboard", "word"),
  excel_path = NULL,
  excel_sheet = NULL,
  clipboard_delim = "\t",
  word_path = NULL,
  verbose = FALSE,
  user_na = TRUE,
  style = NULL
)

Arguments

data

A data.frame.

select

Columns to include. If regex = FALSE, use tidyselect syntax or a character vector of column names (default: tidyselect::everything()). If regex = TRUE, provide a regular expression pattern (character string). The by and weights columns are not part of the selection: everything() and the other helpers skip them, and naming one in select, or giving its position, is an error. A name that is not a column of data is an error (class spicy_missing_column), and so is a selection with no numeric column.

by

Optional grouping column. Accepts an unquoted column name or a single character column name. Coerced to factor for grouping; non-numeric grouping columns (factor, character, logical) are supported as-is. Factor levels keep their declared order, which is the way to choose another order; any other by is sorted, as in table_categorical(): text in the C locale (capitals before lower case, the same on every machine), FALSE before TRUE, numbers ascending, whatever the order of the rows. The group of missing values comes last. A haven labelled by names its groups by their value labels, as the rows of a labelled variable of table_categorical() name its levels ("[1] Women"), in the order of its labels attribute when every observed value carries a label and by ascending code otherwise; declared missing values follow user_na as usual.

exclude

Columns to exclude. Supports tidyselect syntax and character vectors of column names.

regex

Logical. If FALSE (the default), uses tidyselect helpers. If TRUE, the select argument is treated as a regular expression.

drop_na

Logical. Controls how missing values in the by column are handled, under the rule of the family: a table that counts shows the missing values by default (a frequency table reports how many answers are missing, as a "(Missing)" row and, under by, a "(Missing)" column: table_categorical()); a table that computes removes them by default and counts them in the note. This table computes: a mean is computed on the observed values, so the NAs of each summarized variable are always excluded from that variable's statistics, and the exclusion is disclosed in a table note ("Missing values removed: ...") rather than silent. If TRUE (the default), rows with NA in by are removed from the grouped summaries and counted in the same note, by after the variables ("Missing values removed: age (3), sex (2)."): a missing by takes the row out of the whole table, a missing value only out of its own variable's statistics. If FALSE, rows with NA in by form a dedicated "(Missing)" group – the field convention for descriptive tables (gtsummary's "Unknown" row; see the Epidemiologist R Handbook, Descriptive tables) – while the group-comparison test and effect size are still computed on the observed groups only (show the missing, test the observed, matching table_categorical()); a level of by displayed with the label of that group is an error, as there. Under drop_na = TRUE, a by that no observation holds is an error (class spicy_invalid_data). Ignored (with a warning) when by is not used.

weights

Optional case weights: an unquoted column name, a character column name, or a numeric vector of length nrow(data). Weights must be non-negative and finite; rows with NA or zero weight leave every statistic (including Min / Max) and NA weights are disclosed in the table note. The weighted formulas follow the frequency-expansion convention – see the Weights section. The weighted table names its weights in the note ("Statistics weighted by ..."). Group tests and effect sizes are not computed under weights: use table_continuous_lm() for weighted comparisons.

rescale

Logical. If TRUE, weights are first normalized so that they sum to the number of observations used for each variable – the same rescale grammar as table_categorical(). This is the sampling-weights reading: results become invariant to the scale of the weights, and the weighted SD then equals Stata's [aweight] / survey::svyvar() value exactly. FALSE uses the weights as given (frequency reading). The default reads options(spicy.rescale), FALSE when it is unset.

test

Character. Statistical test to use when comparing groups. One of "welch" (default), "student", or "nonparametric".

  • "welch": Welch t-test (2 groups) or Welch one-way ANOVA (3+ groups). Does not assume equal variances.

  • "student": Student t-test (2 groups) or classic one-way ANOVA (3+ groups). Assumes equal variances.

  • "nonparametric": Wilcoxon rank-sum / Mann–Whitney U (2 groups) or Kruskal–Wallis H (3+ groups).

Used whenever by is supplied (since p_value defaults to TRUE in that case) or when statistic = TRUE or an effect_size is requested. Ignored when by is not used, or when all three display toggles are turned off. A variable with no variance has no test, whichever is chosen, with a warning (class spicy_undefined_stat); so has a variable whose groups are each constant on their own value (a residual variance of zero) under "welch" and "student", while the rank tests, which read the order of the values, run. Its test cells (test, p-value, effect size) show the undefined mark, as do those of any comparison that cannot run (too few observations in a group, a test that fails, a non-finite effect size).

p_value

Logical or NULL. If TRUE and by is used, adds a p-value column from the test specified by test. When NULL (the default), the p-value is shown automatically whenever by is supplied, and hidden otherwise. Pass p_value = FALSE to suppress the column explicitly. Ignored when by is not used.

p_adjust

Multiple-comparison adjustment of the p-value column, applied to the family of every test in the table (one p-value per variable). One of "none" (default), "holm", "hochberg", "hommel", "bonferroni", "BH" / "fdr", or "BY", as in table_regression(); delegated to stats::p.adjust(). Needs by and the p-value column. See Multiple-comparison adjustment.

statistic

Logical. If TRUE and by is used, the test statistic is shown in an additional column (e.g., t(df) = ..., F(df1, df2) = ..., W = ..., or H(df) = ...). Both p_value and statistic are independent; either or both can be enabled. Defaults to FALSE. Ignored when by is not used.

show_n

Logical. If TRUE, includes an unweighted n column in the printed ASCII table and in every rendered output (tinytable, gt, flextable, word, excel, clipboard). Set to FALSE to drop the n column structurally from those outputs (no empty placeholder, no spanner). The n column is always present in output = "long" for downstream programmatic access. Defaults to TRUE. Ignored (with a warning) when show_columns is supplied.

show_columns

Statistics to display, as a character vector of tokens or a named list of such vectors (one per variable). NULL (the default) keeps the historical display: mean, SD, min, max, the mean CI (see ci) and n (see show_n). See the "Choosing the statistics" section for the token vocabulary, the per-variable form, and the test that follows a median.

effect_size

Effect-size measure to include in the rendered outputs. One of:

  • "none" (default): no effect-size column.

  • "auto": auto-select the canonical measure for the chosen test and group count – Hedges's g (parametric, 2 groups), eta squared (parametric, 3+ groups), rank-biserial r (nonparametric, 2 groups), epsilon squared (nonparametric, 3+ groups).

  • "hedges_g": Hedges's g (bias-corrected standardized mean difference, 2 groups, parametric). CI via the Hedges & Olkin normal approximation.

  • "eta2": Eta squared (\(\eta^2\), parametric ANOVA-style SS_between / SS_total). CI via inversion of the noncentral F distribution.

  • "r_rb": Rank-biserial r from the Wilcoxon / Mann-Whitney statistic (2 groups, nonparametric). CI via Fisher z-transform.

  • "epsilon2": Epsilon squared (\(\varepsilon^2\)) from the Kruskal-Wallis statistic (3+ groups, nonparametric). CI via percentile bootstrap (2 000 replicates).

A squared measure ends in 2, as in table_continuous_lm() and table_regression(): "eta_sq" and "epsilon_sq", the spellings before 0.14.0, are an error (class spicy_defunct) that gives the new one. A logical value is an error of the same class: use "auto" for TRUE and "none" for FALSE. Explicit choices are validated against the active test and the number of groups; an incompatible request (e.g. "eta2" with two groups, or "hedges_g" with test = "nonparametric") triggers an actionable error. Ignored when by is not used.

effect_size_ci

Logical. If TRUE, appends the confidence interval of the effect size in brackets (e.g., g = 0.45 [0.22, 0.68]), under a header that states its level (ES [95% CI]). Implies a non-"none" effect size: if left at the default effect_size = "none", the function warns and promotes effect_size to "auto" so the requested CI can be shown. Defaults to FALSE.

smd

Logical. If TRUE, adds an SMD column holding the standardized mean difference between the two groups of by, the balance diagnostic of the Table 1 literature. Requires exactly two groups; signed, group 1 minus group 2 in the order the table displays them; no confidence interval and no p-value, by design. It is independent of p_value and of effect_size: turning it on turns nothing else off. Rounded with effect_size_digits. See the "Standardized mean difference" section below. Defaults to FALSE.

ci

Logical. If TRUE, includes the mean confidence interval columns (<level>% CI LL / <level>% CI UL) and their spanner in the printed ASCII table and in every rendered output (tinytable, gt, flextable, word, excel, clipboard). Set to FALSE to drop both columns and the CI spanner structurally from those outputs (no empty placeholders, no border lines under an empty header). The CI bounds are always present as ci_lower / ci_upper in output = "long" for downstream programmatic access. Defaults to TRUE. The CI level is taken from ci_level. Ignored (with a warning) when show_columns is supplied.

labels

How each variable is named in the table: in the rows, in the title, and in the notes. Four forms:

  • labels = FALSE shows the column name.

  • labels = TRUE shows the column's label attribute (as set by haven, labelled, or label_from_names()) when it is present, not NA, and not empty, and the column name otherwise.

  • labels = c(bmi = "Body mass index"), a named character vector, gives the columns it names these labels. Every other column follows options(spicy.var_labels): its name when the option is unset or FALSE, its label attribute when it is TRUE.

  • labels = var_labels(bmi = "BMI (kg/m2)") gives the columns it names these labels, and every other column its label attribute, whatever the option says. var_labels() alone is labels = TRUE.

A name that matches no column of the table warns (class spicy_unknown_name) and is left out, so one vector can still serve every table of a project. The default reads options(spicy.var_labels), and a TRUE or FALSE typed in the call wins over it. Any other value, NULL included, is an error. A variable label documents the data file; a published table carries the wording you set here. note_labels gives a variable a shorter label in the title and the notes.

note_labels

The label a variable takes where the table names it in a sentence, in the title and in the notes, in place of the one labels gives it: a named character vector, note_labels = c(sex = "sex at birth"), or a var_labels() object, which reads as one. labels can then name a row with a whole question while the note says "Missing values removed: sex at birth (2).". The rows, the column headers and the spanners keep the labels of labels, and so do the title and the notes for a variable note_labels does not name. A name that matches no column warns and is left out, as in labels, and NULL (the default) renames nothing. A note label that another variable of the table shows is an error (class spicy_label_collision): choose another one, or relabel the other variable.

ci_level

Confidence level for the mean confidence interval (default: 0.95): a number strictly between 0 and 1.

digits

Number of decimal places of the statistics on each variable's own scale: the mean, SD, median, IQR, quartiles, minimum and maximum, and the confidence intervals of the mean and of the median. One non-negative whole number sets them for every variable (default: 2), and the weighted n with them. A vector named by column, not by label, gives each variable it names its own, digits = c(height = 0, bmi = 1); the other variables and the weighted n keep the default, the digits of a spicy_style() in force included. A name that matches no variable warns and is left out, as in labels. A fraction is an error, as for every *_digits argument. The other numbers keep their own decimals: p-values (p_digits), effect sizes and the SMD (effect_size_digits), test statistics (two decimals) and counts (none).

effect_size_digits

Number of decimal places for effect-size values in formatted displays (default: 2).

p_digits

Integer >= 1. Number of decimal places used to render p-values in the p column (default: 3, the APA Publication Manual standard). Both the displayed precision and the small-p threshold derive from this argument: p_digits = 3 prints .045 and <.001; p_digits = 4 prints .0451 and <.0001; p_digits = 2 prints .05 and <.01. Useful for genomics / GWAS contexts with very small p-values, or for journals using a coarser convention. Leading zeros are always stripped, following APA convention.

decimal_mark

The decimal mark, a single character: "." (the default) or "," in most styles; the Lancet style of spicy_style() uses the middle dot.

align

Horizontal alignment of numeric columns in the printed ASCII table and in the tinytable, gt, flextable, word, excel, and clipboard outputs. The first column (Variable) and Group (when present) are always left-aligned. One of:

  • "decimal" (default): align numeric columns on the decimal mark, the standard scientific-publication convention used by SPSS, SAS, and LaTeX siunitx. Numeric cells are pre-padded with figure-spaces (U+2007, digit-width) so every string in a column has the same width with the decimal mark at the same internal position; centering those uniform-width strings then stacks the decimal points vertically. The same pad-then-center strategy is applied on every rendering engine (gt, tinytable, flextable, word, ASCII print) for a homogeneous rendering, matching table_regression() and table_continuous_lm(). The clipboard output is delimited text meant to be parsed rather than read at a fixed width, so its cells travel unpadded (a padded number pastes as text next to an unpadded number).

  • "center": center-align all numeric columns.

  • "right": right-align all numeric columns.

"center" and "right" reach the excel output too. "decimal" does not: Excel cells are written unpadded, because cell-string padding does not align decimals under a proportional font, so the workbook keeps the engine's own convention instead – counts and the p-value right-aligned, the other numeric columns centered. Same default and same three values as table_continuous_lm(), whose excel output still uses that convention at every align.

note

The note under the table. NULL (the default) builds it: the general note, one paragraph opened by the lead-in of the style ("Note." by default, none under the jama and lancet themes; see note_label in spicy_style()), then, in a table_regression() with stars, the legend of the stars on a line of its own. A single string replaces the general note, the sentences the table writes about its data included. It is printed as typed, line breaks ("\n") included, after the lead-in of the style, which a string opening with a lead-in of its own does not repeat (note = "Note. Data weighted."), and the legend of the stars stays below it. "" keeps that legend alone, and no note at all without stars. FALSE removes the note from every output, the legend included.

note_add

Sentences to add to the note: a character vector of one or more strings, each printed as typed (no period is added) and joined to the note by a space. They close its paragraph, after the definitions of the abbreviations and before the legend of the stars, in every output and in attr(x, "note"), and the sentences the table writes stay. With a string in note, they follow it, so a wrapper can add a house sentence to every table while one table replaces its note. An NA or an empty string is an error, and so is note_add with note = FALSE, which leaves no note to add to.

output

Output format. One of:

  • "default": an ASCII table object, printed when the call is bare.

  • "data.frame": the displayed table with its numbers, one row per printed row.

  • "long": one row per (variable x group) (or one row per variable when by is not used), with every statistic the table computes.

  • "tinytable" (requires tinytable)

  • "gt" (requires gt)

  • "flextable" (requires flextable)

  • "excel" (requires openxlsx2)

  • "clipboard" (requires clipr)

  • "word" (requires flextable and officer)

excel_path

File path for output = "excel", ending in ".xlsx", in a directory that exists.

excel_sheet

Sheet name for output = "excel". NULL (the default) uses "Descriptives".

clipboard_delim

Delimiter for output = "clipboard" (default: "\t"). A cell holding the delimiter itself, a double quote or a line break is quoted RFC 4180-style, so the grid survives whatever delimiter you choose.

word_path

File path for output = "word", ending in ".docx", in a directory that exists.

verbose

Logical. If TRUE, a message (class spicy_info) names the non-numeric columns left out (default: FALSE).

user_na

Logical. If TRUE (the default), declared missing values never reach the numeric summaries: they are excluded like NA and disclosed in the table note (Declared missing values removed: ...); declared-missing by values form no group. If FALSE, the declared codes are summarized as ordinary numbers. See the "Declared missing values" section of freq().

style

A journal style: a theme name ("jama", "nejm", "lancet", "annals", "apa", "aer"), a spicy_style() object, or NULL (the default). A style only changes DEFAULTS – any argument you pass explicitly wins over it. Set options(spicy.style = ) for document-wide scope. A theme covers numeric formatting, the word that opens the note and the abbreviations the note defines, not full editorial conformity; ?spicy_style lists the exact rules each one encodes and the official document they come from. An unknown name is an error.

Value

Depends on output:

  • "default": a data.frame of class "spicy_continuous_table" / "spicy_table" carrying the rendering metadata as attributes. Its columns are internal and can change without notice: read the numbers from "data.frame", "long", broom::tidy(), broom::glance() or as_structured(). The object is returned visibly, so a bare table_continuous(...) call auto-prints the styled ASCII table at the console while t <- table_continuous(...) stays silent (print t to display the table). The object can be re-coerced via as.data.frame.spicy_continuous_table() or piped into broom::tidy() / broom::glance().

  • "data.frame": the displayed table with its numbers, what as.data.frame() of the default object returns: one row per printed row, the key columns variable (the column name), label (the label displayed for it) and, with by, group and .row_role ("group", or "missing" for the group of missing by values), then one column per displayed column, in the order of the table ("M", "SD", "Min", "Max", "95% CI LL" and "95% CI UL", the bounds printed under 95% CI, "n", "Test", "p", "ES", "SMD", the columns show_columns asks for). A cell that prints several numbers keeps the first in its column and the others beside it under their names in the long output: df and df.residual after "Test", es_ci_lower and es_ci_upper after "ES" when its interval is printed, q1 and q3 after "Mdn [Q1, Q3]".

  • "long": a plain data.frame with one row per (variable x group) and columns variable, label, group and .group_role (the role of group: "group", or "missing" for the group of missing by values) when by is used, mean, sd, min, max, ci_lower, ci_upper, median, q1, q3, iqr, med_ci_lower, med_ci_upper, n, weighted_n. Every statistic is computed whatever show_columns displays. When by is used, the comparison columns follow (populated on the first row of each variable block only):

    • test_type – test identifier (e.g., "welch_t", "welch_anova", "student_t", "anova", "wilcoxon", "kruskal"), with statistic, df and df.residual (the degrees of freedom of a t statistic in df, those of an F statistic in df and df.residual, as broom::glance() names them), when the test ran (for p_value, statistic or an effect size).

    • p.value – with the p-value only, and p_unadjusted under p_adjust.

    • es_type – effect-size identifier ("hedges_g", "eta2", "r_rb", or "epsilon2"), es_value, es_ci_lower, es_ci_upper – the effect size and its interval.

    • smd_type, smd_value – the standardized mean difference; smd_type names the kernel the value came from: "continuous" here.

    The effect-size and SMD columns are there whether or not they were requested, NA when not, so the schema a pipeline indexes into does not move with an argument (the weighted_n rule); the test columns follow p_value.

  • "tinytable": a tinytable object.

  • "gt": a gt_tbl object.

  • "flextable": a flextable object.

  • "excel" / "word": writes to disk and returns the file path invisibly.

  • "clipboard": copies the table and returns the display data.frame invisibly.

The note opens with the sample size, as in table_categorical(): "N = 159." or, with by, "N = 159: Women (n = 80), Men (n = 79).", the rows the table describes and those of each group (the group of missing by values included), a row missing by under drop_na = TRUE or a weight left out, rows rather than weights under weights ("N = 60 (unweighted)."), a row of zero weight left out as it is left out of every statistic, the complete cases of each variable in its n column.

The missing-value disclosure (values excluded from the summaries, and rows removed for a missing by value under drop_na = TRUE) travels with the table on every route, not just the console: "default" prints it under the ASCII table, "tinytable" / "gt" / "flextable" / "word" carry it as a table note, "excel" writes it below the body, and the default object and the data frames of "data.frame" / "long" keep the whole note as printed, its lead-in included, in attr(x, "note") (NULL when there is nothing to say), so a pipeline that renders the numbers itself can still state what left the table. The default object also keeps the title it prints in attr(x, "title"). On the "tinytable" route the note is set one size down; options(spicy.note_style) governs that (see table_regression()).

The Excel sheet carries the same title the console prints on its first row; the table itself starts on row 3.

Choosing the statistics

show_columns selects which statistics the table displays. The tokens, and the column each one produces:

TokenColumnStatistic
"m"Mmean
"sd"SDstandard deviation
"med"Mdnmedian (stats::median())
"iqr"IQRinterquartile width, Q3 - Q1
"med_iqr"Mdn [Q1, Q3]median and the interquartile interval, in one compact column
"q1" / "q3"Q1 / Q3first / third quartile
"min" / "max"Min / Maxextremes
"ci"<level>% CI LL / ULt confidence interval of the mean
"med_ci"Mdn <level>% CI LL / ULexact confidence interval of the median
"n"nvalid observations
"weighted_n"Weighted nsum of weights (requires weights)

The column is the header the table prints. In output = "data.frame", as.data.frame() and as_structured(), the columns are named by their keys, the same in every display language and style: the default header in English (M; Mdn, Mdn [Q1, Q3], Mdn <level>% CI LL / UL for the median), where the "jama" and "lancet" styles print Mean and Median. The note under the table defines the abbreviations these headers show, as the style in force asks (see abbreviations in spicy_style()).

Quartiles use stats::quantile()'s default type 7. "iqr" is the width (one number, the rank mirror of SD); "med_iqr" shows the interval with its bounds.

A character vector is displayed in the order it is written, so c("med", "q1", "q3", "m", "sd", "n") puts the median first. A token written twice counts once, where it first appears. An interval comes after the statistic it bounds, "ci" after "m" and "med_ci" after "med" or "med_iqr": written before it, it is an error. The per-variable list form below has no written order common to all the variables, so its columns follow the order of the table above, which is also the order of the default display.

Weights

With weights, every displayed statistic uses the frequency-expansion convention: for integer weights each statistic equals its unweighted version computed on the data with every row repeated w times (rep(x, w)), exactly; with all weights equal to 1 every statistic equals its unweighted sibling. The formulas, with \(W = \sum w_i\):

  • mean: \(\sum w_i x_i / W\);

  • SD: \(\sqrt{\sum w_i (x_i - \bar{x}_w)^2 / (W - 1)}\);

  • quantiles: type-7 positions on the cumulative-weight scale (the Hmisc::wtd.quantile() default algorithm);

  • CI of the mean: \(\bar{x}_w \pm t_{W-1} \, s_w / \sqrt{W}\);

  • n counts the rows used; "weighted_n" reports \(W\).

These are the conventions of Hmisc::wtd.mean() / wtd.var() / wtd.quantile() (defaults), matrixStats::weightedSd(), and DescTools::Quantile(), and – for integer weights – of Stata's [fweight] and SPSS's WEIGHT BY. With rescale = TRUE the weights are normalized to sum to the number of observations first, which makes every result invariant to the scale of the weights and makes the SD equal Stata's [aweight] / survey::svyvar() value – the reading appropriate for sampling weights. Weighted-quantile conventions genuinely differ across software (Stata interpolates nowhere, SAS refuses analytic-weighted quantiles, the survey package offers twelve rules); spicy states its rule here rather than leaving it implicit.

Two deliberate refusals: the "med_ci" token (an order-statistic interval with no weighted version) and, under by, the group tests and effect sizes – a t-test printed next to weighted descriptives would silently be unweighted. Set p_value = FALSE for weighted descriptives by group, or use table_continuous_lm() with weights for weighted comparisons. Note that table_continuous_lm()'s residual SD answers a different question (model-based, precision-weight convention) and is not expected to match the descriptive SD here.

A named list applies a different selection to each variable, with .default covering the variables it does not name – the case of a table where a skewed variable must be reported as a median while the others keep the mean:

show_columns = list(
  mvpa    = c("med_iqr", "n"),
  sitting = c("med_iqr", "n"),
  .default = c("m", "sd", "n")
)

A name that is no variable of the table warns and is left out, as in labels.

The table's columns are the union of the requested tokens; a cell of a column the variable did not ask for is left blank (structurally empty, not an en dash, which is reserved for an undefined statistic).

The table tests what it shows. A variable displaying a median without a mean takes the rank-based test – Wilcoxon rank-sum for two groups, Kruskal-Wallis beyond – and the rank effect size (rank-biserial r, \(\varepsilon^2\)) when effect_size is "auto". The switch is per variable, so a mixed table carries a rank test on its median rows and Welch on its mean rows, and the table note names which test each variable carries. An explicit test is sovereign: it applies to every variable, with a warning naming the ones displayed as medians.

"med_ci" is the exact order-statistic (sign-test) confidence interval: the tightest interval \([x_{(k)}, x_{(n-k+1)}]\) whose binomial coverage still reaches ci_level. It is distribution-free and deterministic – no bootstrap, no seed – and its coverage is at least nominal, the same convention as SAS PROC UNIVARIATE (CIPCTLDF) and DescTools::MedianCI(method = "exact"). Below about six observations no interval reaches the requested level; the cells then show an en dash rather than a false interval.

"ci" is the confidence interval of the mean: requested without "m" it is dropped with a warning pointing at "med_ci", and "med_ci" without a displayed median is dropped likewise. When show_columns is supplied it decides the n and CI columns on its own, and a contradictory show_n / ci is reported.

Tests

The omnibus test is computed only when by is supplied and at least two groups remain after dropping NAs, with every group contributing at least two observations. Choice of test family is driven by test (see the @param entry for the full dispatch and the underlying stats:: functions called). Whenever the table shows p-values, its note names the test behind them ("P-values: Welch t-test.", "P-value:" when the table shows one), each variable beside its own test when they differ. A test shown by its statistic alone (p_value = FALSE, statistic = TRUE) is named as the group comparison ("Group comparison: Welch t-test.").

No data in a group

A variable with no observed value in a group of by (every value missing there, typically a question not asked in that group) has no data in it, which is not a missing answer. Its cells in that group show an en dash, n included (NA in output = "long" and the typed view), its rows there are not counted in the missing values of the note, the comparison and the effect size use the groups with data (with fewer than two, there is no test and its cell is left blank), and the note says so ("No data for X and Y in G.", X and Y the labels of the variables with no data in group G). Without by, a variable with no observed value at all is shown the same way ("No data for X."). A declared level of by that no row holds keeps its count of zero, and the group of missing by values is never such a group.

For model-based contrasts (heteroskedasticity-consistent SE, cluster-robust SE, weighted contrasts, fitted means, covariate adjustment), use table_continuous_lm().

Effect sizes

See @param effect_size for the dispatch table (canonical measure for each (test, n_groups) combination) and the validation rules applied to explicit requests.

Confidence intervals (enabled with effect_size_ci = TRUE) use noncentral F inversion for \(\eta^2\), the Hedges-Olkin normal approximation for g, the Fisher z-transform for r, and percentile bootstrap (2,000 replicates) for \(\varepsilon^2\). The bootstrap bounds depend on the random number generator state: call set.seed() before the table for reproducible \(\varepsilon^2\) intervals (the other three CIs are closed-form and deterministic).

For Cohen's d, Hays's \(\omega^2\), and Cohen's f\(^2\) (derived from a fitted, possibly weighted lm()), use the model-based companion table_continuous_lm().

Standardized mean difference

smd = TRUE adds an SMD column with the balance diagnostic of the Table 1 literature, in Austin's form (Austin 2009, Stat Med 28:3083-3107; Austin 2011, Multivar Behav Res 46:399-424):

$$\mathrm{SMD} = \frac{\bar{x}_1 - \bar{x}_2}{\sqrt{(s_1^2 + s_2^2) / 2}}$$

The denominator is the root mean of the two group variances, each at \(n - 1\), not the degrees-of-freedom pooled SD. At equal group sizes those two denominators are the same, so the SMD is exactly Cohen's d; at unequal sizes they part company (on a 4-versus-3 split the SMD is \(-0.51\) and d is \(-0.54\)).

effect_size = "hedges_g", two columns to the left, is a third number: g applies the small-sample correction J on top of d, so it never equals the SMD. At equal group sizes the ratio \(g / \mathrm{SMD}\) is exactly J – 0.80 at n = 3 per group, 0.96 at n = 10 – approaching 1 only as the sample grows. Read each for what it is; do not recompute one from the other. (The divergence is nameable upstream: cobalt::col_w_smd(s.d.denom = "pooled") reproduces this column, s.d.denom = "hedges" reproduces hedges_g.)

Conventions, all deliberate:

  • Signed, group 1 minus group 2 in the order the table displays the groups – the two groups sit side by side, so a bare magnitude would make the reader re-derive a direction the row already gives. (tableone publishes the magnitude; cobalt and arsenal sign it the other way, guessing the second level as "treated".) The threshold in the table note is read on \(|\mathrm{SMD}|\); the column keeps the sign. No conditional formatting: spicy never highlights a threshold.

  • No confidence interval and no p-value, ever. The SMD is a descriptive diagnostic; attaching an interval to it reintroduces the test reasoning the balance literature asks the reader to drop. This is not a missing feature.

  • Exactly two groups. A by with three or more is refused rather than averaged over pairs: an average has no published reading, and it can sit under the usual threshold while one pair sits well over it.

  • Complete cases on the observed groups. A drop_na = FALSE "(Missing)" group is displayed and never enters the diagnostic, exactly as the test and the effect size behave.

  • Independent of p_value. Turning the SMD on turns nothing else off. The balance-table idiom is smd = TRUE, p_value = FALSE; you have to write both.

Under weights, the means and variances are the weighted ones the M and SD columns already display – the frequency convention of the Weights section, from the same producer, so the column cannot contradict its neighbors. One consequence follows and is intended: a frequency weight is a number of copies, so the weighted SMD is not invariant to the scale of the weights (multiplying every weight by ten moves it, as it moves the SD column). rescale = TRUE normalizes the weights to sum to n, restores scale invariance, and is the form to use for sampling weights until the dedicated survey-design functions land.

A cell is an en-dash when the diagnostic applies but cannot be estimated: both groups constant at different values (an infinite standardized distance, disclosed by a warning), or a group with too little data to have a variance (silent – the SD cell beside it already says so). Two groups constant at the same value are perfectly balanced and print 0.00.

Display conventions

Decimal alignment, p-value formatting, and required suggested packages per output engine are documented under @param align, @param p_digits, and @param output respectively.

Non-numeric columns are silently dropped (set verbose = TRUE to see which columns were excluded). When a constant column is passed, its statistics are reported exactly: SD is 0.00 and the CI degenerates to [m, m]. An en-dash cell appears only when a statistic is undefined (fewer than two valid observations).

Declared missing values

Survey files imported with haven often carry declared missing values: codes such as 8 = Don't know or 9 = Refused that the source file marks as missing while keeping them distinct from a plain NA. Two kinds of declaration exist: na_values / na_range metadata on haven::labelled_spss() vectors, and tagged missing values created by haven::tagged_na() (the Stata .a, .b, ... convention). haven::read_sav() keeps the na_values / na_range declaration only with user_na = TRUE: its default turns the declared codes into NA on import.

spicy honors the declaration by default (user_na = TRUE): declared missing values are excluded from every statistic exactly like NA – valid percentages, means, chi-squared tests, association measures, row-wise summaries, and group definitions – but they are not erased from display. freq() lists each observed declared value as its own row of the Missing block, with its value label; cross_tab(), table_categorical(), and table_continuous() disclose the exclusion in the table note (Declared missing values removed: x (2).); varlist() counts them as missing in N_valid / NAs / N_distinct while still listing the declared codes in Values; code_book() counts them in n_declared_missing and flags them in its values table.

Every function involved offers the same escape hatch: set user_na = FALSE to ignore the declaration and treat the declared codes as valid values (the behavior of spicy before 0.13.0). Tagged missing values are genuine NAs either way; for them, user_na = FALSE only collapses the per-tag breakdown back into the regular NA count.

Multiple-comparison adjustment

The p-values of a descriptive table describe: they test no prespecified hypothesis. STROBE (explanation of item 14) and the SAMPL guidelines advise against significance tests in descriptive and baseline tables, hence the default p_adjust = "none".

When a journal or a protocol asks for adjusted p-values anyway, p_adjust applies stats::p.adjust() to the family of every test in the table: one p-value per variable (one per block in table_outcome()), never the rows of one variable, which share one test. "holm" suits a confirmatory set of comparisons, "BH" a screening one; table_regression() lists the methods. A test that returned no p-value stays out of the family.

The adjustment runs before any formatting, so the p column and its <.001 threshold show the adjusted values, and the table note gives the number of tests and the method. The raw p-values stay in the object: a p_unadjusted column next to the p-value column of the returned data frames, and col_meta$p$p_unadjusted (with the method in p_adjust and the family size in p_adjust_m) in as_structured(). Without by, or with p_value = FALSE, the table shows no p-value to adjust, and a p_adjust other than "none" is an error.

See also

table_outcome() for the transposed shape – ONE continuous outcome across the levels of SEVERAL groupings, one block of rows per grouping. Several outcomes across one grouping is this function; one outcome across one or more groupings is that one; table_continuous_lm() for the model-based companion (heteroskedasticity-consistent SE, cluster-robust SE, weighted contrasts, fitted means); table_categorical() for categorical variables; freq() for one-way frequency tables; cross_tab() for two-way cross-tabulations.

Other spicy tables: table_categorical(), table_continuous_lm()

Examples

# --- Basic usage ---------------------------------------------------------

# Default: ASCII console table.
table_continuous(
  sochealth,
  select = c(bmi, wellbeing_score)
)
#> Descriptive statistics
#> 
#>                                                    95% CI          
#>                                                 ────────────       
#>  Variable        │   M     SD     Min    Max     LL     UL     n   
#> ─────────────────┼─────────────────────────────────────────────────
#>  bmi             │ 25.93   3.72  16.00   38.90  25.72  26.14  1188 
#>  wellbeing_score │ 69.04  15.62  18.70  100.00  68.16  69.93  1200 
#> 
#> Note. N = 1200. Missing values removed: bmi (12). CI = confidence interval; LL = lower limit; UL = upper limit.

# Grouped by education (Welch p-value added by default).
table_continuous(
  sochealth,
  select = c(bmi, wellbeing_score),
  by = education
)
#> Descriptive statistics by education
#> 
#>                                                                     95% CI    
#>                                                                  ──────────── 
#>  Variable        │ Group              M     SD     Min    Max     LL     UL   
#> ─────────────────┼────────────────────────────────────────────────────────────
#>  bmi             │ Lower secondary  28.09   3.47  18.20   38.90  27.66  28.51 
#>                  │ Upper secondary  26.02   3.43  16.00   37.10  25.73  26.31 
#>                  │ Tertiary         24.39   3.52  16.00   33.00  24.04  24.74 
#> ╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌┼╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌
#>  wellbeing_score │ Lower secondary  57.22  15.44  18.70   97.90  55.33  59.10 
#>                  │ Upper secondary  68.97  13.62  26.70  100.00  67.82  70.12 
#>                  │ Tertiary         76.85  13.23  40.40  100.00  75.55  78.15 
#> 
#>  Variable        │ Group             n     p   
#> ─────────────────┼─────────────────────────────
#>  bmi             │ Lower secondary  260  <.001 
#>                  │ Upper secondary  534        
#>                  │ Tertiary         394        
#> ╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌┼╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌
#>  wellbeing_score │ Lower secondary  261  <.001 
#>                  │ Upper secondary  539        
#>                  │ Tertiary         400        
#> 
#> Note. N = 1200: Lower secondary (n = 261), Upper secondary (n = 539), Tertiary (n = 400). Missing values removed: bmi (12). P-values: Welch one-way ANOVA. CI = confidence interval; LL = lower limit; UL = upper limit.

# Test statistic alongside the p-value.
table_continuous(
  sochealth,
  select = c(bmi, wellbeing_score),
  by = education,
  statistic = TRUE
)
#> Descriptive statistics by education
#> 
#>                                                                     95% CI    
#>                                                                  ──────────── 
#>  Variable        │ Group              M     SD     Min    Max     LL     UL   
#> ─────────────────┼────────────────────────────────────────────────────────────
#>  bmi             │ Lower secondary  28.09   3.47  18.20   38.90  27.66  28.51 
#>                  │ Upper secondary  26.02   3.43  16.00   37.10  25.73  26.31 
#>                  │ Tertiary         24.39   3.52  16.00   33.00  24.04  24.74 
#> ╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌┼╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌
#>  wellbeing_score │ Lower secondary  57.22  15.44  18.70   97.90  55.33  59.10 
#>                  │ Upper secondary  68.97  13.62  26.70  100.00  67.82  70.12 
#>                  │ Tertiary         76.85  13.23  40.40  100.00  75.55  78.15 
#> 
#>  Variable        │ Group             n           Test             p   
#> ─────────────────┼────────────────────────────────────────────────────
#>  bmi             │ Lower secondary  260  F(2, 654.48) = 87.96   <.001 
#>                  │ Upper secondary  534                               
#>                  │ Tertiary         394                               
#> ╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌┼╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌
#>  wellbeing_score │ Lower secondary  261  F(2, 638.59) = 144.35  <.001 
#>                  │ Upper secondary  539                               
#>                  │ Tertiary         400                               
#> 
#> Note. N = 1200: Lower secondary (n = 261), Upper secondary (n = 539), Tertiary (n = 400). Missing values removed: bmi (12). P-values: Welch one-way ANOVA. CI = confidence interval; LL = lower limit; UL = upper limit.

# --- Choosing the statistics --------------------------------------------

# Median and interquartile range instead of mean and SD.
table_continuous(
  sochealth,
  select = c(bmi, wellbeing_score),
  show_columns = c("med_iqr", "n")
)
#> Descriptive statistics
#> 
#>  Variable          │      Mdn [Q1, Q3]         n    
#> ───────────────────┼────────────────────────────────
#>  bmi               │  25.90 [23.40, 28.60]    1188  
#>  wellbeing_score   │  70.25 [58.90, 79.23]    1200  
#> 
#> Note. N = 1200. Missing values removed: bmi (12). Q1 = first quartile; Q3 = third quartile.

# Median with its exact (order-statistic) confidence interval.
table_continuous(
  sochealth,
  select = bmi,
  show_columns = c("med", "iqr", "med_ci", "n")
)
#> Descriptive statistics
#> 
#>                                   Mdn 95% CI            
#>                                ────────────────         
#>  Variable   │   Mdn     IQR      LL       UL       n    
#> ────────────┼───────────────────────────────────────────
#>  bmi        │  25.90    5.20    25.70    26.20    1188  
#> 
#> Note. N = 1200. Missing values removed: bmi (12). Mdn 95% CI = exact order-statistic confidence interval for the median (coverage at least 95%). IQR = interquartile range (Q3 − Q1); CI = confidence interval; LL = lower limit; UL = upper limit.

# One selection per variable. A skewed variable that a scoring
# protocol requires in median and IQR (the IPAQ case) sits next to
# variables kept in mean and SD; each row is tested the way it is
# displayed, and the note says so.
table_continuous(
  sochealth,
  select = c(bmi, life_sat_health, wellbeing_score),
  by = sex,
  show_columns = list(
    life_sat_health = c("med_iqr", "n"),
    .default = c("m", "sd", "n")
  )
)
#> Descriptive statistics by sex
#> 
#>  Variable        │ Group     M     SD      Mdn [Q1, Q3]      n     p   
#> ─────────────────┼─────────────────────────────────────────────────────
#>  bmi             │ Female  25.69   3.78                     616   .018 
#>                  │ Male    26.20   3.64                     572        
#> ╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌┼╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌
#>  life_sat_health │ Female                4.00 [3.00, 5.00]  616   .233 
#>                  │ Male                  4.00 [3.00, 5.00]  576        
#> ╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌┼╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌
#>  wellbeing_score │ Female  67.16  14.80                     620  <.001 
#>                  │ Male    71.05  16.23                     580        
#> 
#> Note. N = 1200: Female (n = 620), Male (n = 580). Missing values removed: bmi (12), life_sat_health (8). P-values: Welch t-test (bmi, wellbeing_score); Wilcoxon rank-sum test (life_sat_health). Q1 = first quartile; Q3 = third quartile.

# --- Effect sizes -------------------------------------------------------

# Auto-selected effect size with confidence interval (Hedges's g for
# binary `by`, eta squared for k > 2).
table_continuous(
  sochealth,
  select = wellbeing_score,
  by = sex,
  effect_size = "auto",
  effect_size_ci = TRUE
)
#> Descriptive statistics by sex
#> 
#>                                                            95% CI         
#>                                                         ────────────      
#>  Variable        │ Group     M     SD     Min    Max     LL     UL     n  
#> ─────────────────┼────────────────────────────────────────────────────────
#>  wellbeing_score │ Female  67.16  14.80  19.60  100.00  65.99  68.33  620 
#>                  │ Male    71.05  16.23  18.70  100.00  69.73  72.37  580 
#> 
#>  Variable        │ Group     p          ES [95% CI]        
#> ─────────────────┼─────────────────────────────────────────
#>  wellbeing_score │ Female  <.001  g = -0.25 [-0.36, -0.14] 
#>                  │ Male                                    
#> 
#> Note. N = 1200: Female (n = 620), Male (n = 580). P-value: Welch t-test. CI = confidence interval; LL = lower limit; UL = upper limit; ES = effect size.

# Explicit effect-size measure.
table_continuous(
  sochealth,
  select = wellbeing_score,
  by = education,
  effect_size = "eta2",
  effect_size_ci = TRUE,
  effect_size_digits = 3
)
#> Descriptive statistics by education
#> 
#>                                                                     95% CI    
#>                                                                  ──────────── 
#>  Variable        │ Group              M     SD     Min    Max     LL     UL   
#> ─────────────────┼────────────────────────────────────────────────────────────
#>  wellbeing_score │ Lower secondary  57.22  15.44  18.70   97.90  55.33  59.10 
#>                  │ Upper secondary  68.97  13.62  26.70  100.00  67.82  70.12 
#>                  │ Tertiary         76.85  13.23  40.40  100.00  75.55  78.15 
#> 
#>  Variable        │ Group             n     p           ES [95% CI]        
#> ─────────────────┼────────────────────────────────────────────────────────
#>  wellbeing_score │ Lower secondary  261  <.001  η² = 0.208 [0.169, 0.246] 
#>                  │ Upper secondary  539                                   
#>                  │ Tertiary         400                                   
#> 
#> Note. N = 1200: Lower secondary (n = 261), Upper secondary (n = 539), Tertiary (n = 400). P-value: Welch one-way ANOVA. CI = confidence interval; LL = lower limit; UL = upper limit; ES = effect size.

# --- Selection helpers --------------------------------------------------

# Regex selection.
table_continuous(
  sochealth,
  select = "^life_sat",
  regex = TRUE
)
#> Descriptive statistics
#> 
#>                                                     95% CI         
#>                                                   ──────────       
#>  Variable               │  M     SD   Min   Max    LL    UL    n   
#> ────────────────────────┼──────────────────────────────────────────
#>  life_sat_health        │ 3.55  1.25  1.00  5.00  3.48  3.62  1192 
#>  life_sat_work          │ 3.38  1.18  1.00  5.00  3.31  3.45  1192 
#>  life_sat_relationships │ 3.72  1.10  1.00  5.00  3.66  3.79  1192 
#>  life_sat_standard      │ 3.40  1.16  1.00  5.00  3.33  3.46  1192 
#> 
#> Note. N = 1200. Missing values removed: life_sat_health (8), life_sat_work (8), life_sat_relationships (8), life_sat_standard (8). CI = confidence interval; LL = lower limit; UL = upper limit.

# Pretty labels keyed by column name.
table_continuous(
  sochealth,
  select = c(bmi, life_sat_health),
  labels = c(
    bmi = "Body mass index",
    life_sat_health = "Satisfaction with health"
  )
)
#> Descriptive statistics
#> 
#>                                                           95% CI          
#>                                                        ────────────       
#>  Variable                 │   M     SD    Min    Max    LL     UL     n   
#> ──────────────────────────┼───────────────────────────────────────────────
#>  Body mass index          │ 25.93  3.72  16.00  38.90  25.72  26.14  1188 
#>  Satisfaction with health │  3.55  1.25   1.00   5.00   3.48   3.62  1192 
#> 
#> Note. N = 1200. Missing values removed: Body mass index (12), Satisfaction with health (8). CI = confidence interval; LL = lower limit; UL = upper limit.

# --- Output formats -----------------------------------------------------

# The rendered outputs below all wrap the same call:
#   table_continuous(sochealth,
#                    select = c(bmi, wellbeing_score),
#                    by = sex)
# only `output` changes. Assign each result to a variable -- some
# engines auto-print as a console-friendly text fallback inside
# the `?` help viewer.

# Wide / long data.frame (synonyms): one row per (variable x group).
table_continuous(
  sochealth,
  select = c(bmi, wellbeing_score),
  by = sex,
  output = "data.frame"
)
#>          variable           label  group .row_role        M        SD  Min
#> 1             bmi             bmi Female     group 25.68506  3.781113 16.0
#> 2             bmi             bmi   Male     group 26.19685  3.638092 16.0
#> 3 wellbeing_score wellbeing_score Female     group 67.16194 14.798488 19.6
#> 4 wellbeing_score wellbeing_score   Male     group 71.04879 16.227304 18.7
#>     Max 95% CI LL 95% CI UL   n            p
#> 1  38.9  25.38588  25.98425 616 1.760093e-02
#> 2  37.7  25.89808  26.49563 572           NA
#> 3 100.0  65.99480  68.32907 620 1.647005e-05
#> 4 100.0  69.72540  72.37219 580           NA

# \donttest{
# Rendered HTML / docx objects -- best viewed inside a
# Quarto / R Markdown document or a pkgdown article.
if (requireNamespace("tinytable", quietly = TRUE)) {
  tt <- table_continuous(
    sochealth, select = c(bmi, wellbeing_score), by = sex,
    output = "tinytable"
  )
}
if (requireNamespace("gt", quietly = TRUE)) {
  tbl <- table_continuous(
    sochealth, select = c(bmi, wellbeing_score), by = sex,
    output = "gt"
  )
}
if (requireNamespace("flextable", quietly = TRUE)) {
  ft <- table_continuous(
    sochealth, select = c(bmi, wellbeing_score), by = sex,
    output = "flextable"
  )
}

# Excel and Word: write to a temporary file.
if (requireNamespace("openxlsx2", quietly = TRUE)) {
  tmp <- tempfile(fileext = ".xlsx")
  table_continuous(
    sochealth, select = c(bmi, wellbeing_score), by = sex,
    output = "excel", excel_path = tmp
  )
  unlink(tmp)
}
if (
  requireNamespace("flextable", quietly = TRUE) &&
    requireNamespace("officer", quietly = TRUE)
) {
  tmp <- tempfile(fileext = ".docx")
  table_continuous(
    sochealth, select = c(bmi, wellbeing_score), by = sex,
    output = "word", word_path = tmp
  )
  unlink(tmp)
}
# }

if (FALSE) { # \dontrun{
# Clipboard: writes to the system clipboard.
table_continuous(
  sochealth, select = c(bmi, wellbeing_score), by = sex,
  output = "clipboard"
)
} # }