crosstable: Easily describe datasets

Description

Generate a descriptive table of all chosen columns, as contingency tables for categorical variables and as calculation summaries for numeric variables. If the by argument points to one or several categorical variables, crosstable will output a description of all columns for each level. Otherwise, if it points to a numeric variable, crosstable will calculate correlation coefficients with all other selected numeric columns. Finally, if it points to a Surv object, crosstable will describe the survival at different times.

Can be formatted as an HTML table using as_flextable().

Usage

crosstable(
  data,
  cols = everything(),
  ...,
  by = NULL,
  total = c("none", "row", "column", "both"),
  percent_pattern = "{n} ({p_row})",
  percent_digits = 2,
  num_digits = 1,
  showNA = c("ifany", "always", "no"),
  label = TRUE,
  funs = c(` ` = cross_summary),
  funs_arg = list(),
  cor_method = c("pearson", "kendall", "spearman"),
  drop_levels = FALSE,
  remove_zero_percent = NULL,
  unique_numeric = 3,
  date_format = NULL,
  times = NULL,
  followup = FALSE,
  test = FALSE,
  test_args = crosstable_test_args(),
  effect = FALSE,
  effect_args = crosstable_effect_args(),
  margin = deprecated(),
  .vars = deprecated()
)

Value

A data.frame/tibble of class crosstable

Arguments

data: A data.frame
cols: <tidy-select> Columns to describe, default to everything(). See examples or vignette("crosstable-selection") for more details.
...: Unused. All parameters after this one must be named.
by: The variable to group on. Character or name.
total: one of ["none", "row", "column" or "both"] to indicate whether to add total rows and/or columns. Default to none.
percent_pattern: Pattern used to describe proportions in categorical data. Syntax uses a glue::glue() specification, see the section below for more details. Default to "{n} ({p_col})" if by is null and "{n} ({p_row})" if it is not.
percent_digits: Number of digits for percentages.
num_digits: Number of digits for numeric summaries.
showNA: Whether to show NA in categorical variables (one of c("ifany", "always", "no"), like in table()).
label: Whether to show labels. See import_labels() or set_label()for how to add labels to the dataset columns.
funs: Functions to apply to numeric variables. Default to cross_summary().
funs_arg: Additional parameters for funs, e.g. digits (the number of decimal places) for the default cross_summary(). Ultimately, these arguments are passed to format_fixed().
cor_method: One of c("pearson", "kendall", "spearman") to indicate which correlation coefficient is to be used.
drop_levels: Whether to drop unused levels of factor variables. Default to TRUE.
remove_zero_percent: Whether to remove proportions when n==0. Default to FALSE.
unique_numeric: The number of non-missing different levels a variable should have to be considered as numeric.
date_format: if x is a vector of Date or POSIXt, the format to apply (see strptime for formats)
times: When using formula with survival::Surv() objects, which times to summarize.
followup: When using formula with survival::Surv() objects, whether to display follow-up time.
test: Whether to perform tests.
test_args: See crosstable_test_args to override default testing behaviour.
effect: Whether to compute a effect measure.
effect_args: See crosstable_effect_args to override default behaviour.
margin: Deprecated in favor of percent_pattern. One of ["row", "column", "cell", "none", or "all"]. Default to row.
.vars: Deprecated in favor of cols.

About <code>percent_pattern</code>

The percent_pattern argument is very powerful but can be difficult to understand at first :

It is usually a single string that uses the glue syntax, where variables are put in curly braces ({x}).
Counts are expressed as {n}, {n_row}, {n_col}, and {n_tot}, and proportions as {p_row}, {p_col}, and {p_cell}, depending on the margin on which they are calculated.
For each variable, a version including missing values in the total is proposed as {n_xxx_na} or {p_xxx_na}.
For each proportion, a confidence interval is also calculated using Wilson score and can be expressed as {p_xxx_inf} and {p_xxx_sup}. See examples for practical applications.
Alternatively, percent_pattern can be a list of characters with names body, total_row, total_col, and total_all to also control the pattern in other parts of the crosstable than the body.

Author

Dan Chaltiel

Examples

Run this code

#whole table
crosstable(iris)
crosstable(mtcars)
crosstable(mtcars2)

#tidyselection, custom functions
library(dplyr)
crosstable(mtcars2, c(ends_with("t"), starts_with("c")), by=vs,
           funs=c(mean, quantile), funs_arg=list(probs=c(.25,.75)))

#margin and totals, multiple by
crosstable(mtcars2, c(disp, cyl), by=c(am, vs),
           margin=c("row", "col"), total = "both")

#predicate selection, correlation, effect calculation
crosstable(mtcars2, where(is.numeric), by=hp, effect=TRUE)

#lambda selection & statistical tests
crosstable(mtcars2, ~is.numeric(.x) && mean(.x)>50, by=vs, test=TRUE)

#Dates
mtcars2$my_date = as.Date(mtcars2$hp , origin="2010-01-01") %>% set_label("Some nonsense date")
crosstable(mtcars2, my_date, by=vs, date_format="%d/%m/%Y")

#Survival data (using formula syntax)
library(survival)
crosstable(aml, Surv(time, status) ~ x, times=c(0,15,30,150), followup=TRUE)

#Patterns
crosstable(mtcars2, vs, by=am, percent_digits=0,
           percent_pattern="{n} ({p_col} / {p_row})")
crosstable(mtcars2, vs, by=am, percent_digits=0,
           percent_pattern="N={n} \np[95%CI] = {p_col} [{p_col_inf}; {p_col_sup}]")
str_high="n>5"; str_lo="n<=5"
crosstable(mtcars2, vs, by=am, percent_digits=0,
           percent_pattern="col={p_col}, row={p_row} ({ifelse(n<5, str_lo, str_high)})")

Run the code above in your browser using DataLab