knitr::opts_chunk$set( comment = "#>" )
Note: At present, carpenter only creates tables that you would typically see as the 'basic characteristics' or 'descriptive statistics' table in most biomedical articles. The plan is to include other common table structures, but they haven't been implemented yet. If you have any suggestions, let me know as an Issue!
In biomedical research, there are certain types of tables that are often included in the article. For instance, some basic statistics between the treatment and control group. Or maybe it is between males and females, before and after an intervention, and so on. Often these tables are a hassle to create and are prone to needing updates based on slight changes in the data or from reviewer comments. carpenter was designed to make creating these tables easily and for them to be easily updated when data changes.
Before going over the code, I should mention a key note and assumption: how the
table is eventually presented is determined by how the data looks when it is
passed into the carpenter functions. A good example is 'before and after' studies, where
researchers may store data where each row is the participant and the remaining
columns being the before and after values of a measure (e.g. Weight_before
and
Weight_after
as two columns). In this case, the data should ideally be
formatted more like this:
ID | Visit | Weight ---|-------|------- 1 | Before | 50 1 | After | 54 2 | Before | 65 2 | After | 70
See how the data is stored in 'long' format. This is the form of data that carpenter was designed to deal with. For more details on what is tidy data see resources at the end.
The underlying design principal for carpenter is that you create an outline of what the table should look like before finally creating the table. This is how carpenters also work: they sketch what the product will look like before actually starting to build anything. There are four 'outlining' functions, several carpenter statistics functions, and one final 'building' function:
outline_table()
starts the outlining process.add_rows()
adds variables to the row or rows with associated descriptive
statistics.stat_*()
type statistic functions to be used with add_rows()
(a list of
available statistics can be found using ?carpenter::table_stats
).renaming()
for customizing the naming of the rows and table headers.build-table()
for finally building the table (uses the
pander
package).These functions are chained together using the wonderful
magrittr %>%
pipe. If you've
never used this package or the pipe, take a look at the vignette on introducing
it. So, let's do some coding:
library(magrittr) library(carpenter) head(iris) outline_table(iris, 'Species')
You'll notice that the outline_table
function returned a tibble
of 0 rows
and 0 columns. That's because we haven't added anything else to the table!
carpenter is waiting for more information. So we add rows by:
outline_table(iris, 'Species') %>% add_rows('Sepal.Length', stat_meanSD)
You see it has now added a row to the table. Adding more rows:
outline_table(iris, 'Species') %>% add_rows(c('Sepal.Length', 'Petal.Length'), stat_meanSD) %>% add_rows('Sepal.Width', stat_medianIQR)
See how this can make creating these tables very easy. Let's remove the .
from
the row names and fix up the table header names:
tab <- outline_table(iris, 'Species') %>% add_rows(c('Sepal.Length', 'Petal.Length'), stat_meanSD) %>% add_rows('Sepal.Width', stat_medianIQR) %>% renaming('header', c('Measures', 'Setosa', 'Versicolor', 'Virginica')) %>% renaming('rows', function(x) gsub('\\.', ' ', x)) tab
And finally build the table into a Markdown format for easy insertion into R Markdown documents:
build_table(tab)
If you have factor/discrete data, you can even use include these variables:
library(dplyr, quietly = TRUE) mtcars %>% mutate( gear = as.factor(gear), vs = as.factor(vs) ) %>% outline_table('vs') %>% add_rows('mpg', stat_meanSD) %>% add_rows('drat', stat_medianIQR) %>% add_rows('gear', stat_nPct) %>% renaming('header', c('Measures', 'V-engine', 'Straight engine')) %>% renaming('rows', function(x) x %>% gsub('drat', 'Read axle ratio', .) %>% gsub('mpg', 'Miles/gallon', .) %>% gsub('gear', 'Number of gears', .)) %>% build_table()
Pretty easy eh?
Sometimes, though, you don't need to compare multiple columns, but instead need
to only show one column. Easy, just don't include a header in the
outline_table()
!
iris %>% outline_table() %>% add_rows('Sepal.Length', stat_meanSD) %>% add_rows('Sepal.Width', stat_medianIQR) %>% renaming('header', c('Measures', 'Values')) %>% build_table()
If you don't use R Markdown, you can continue the chain into write.csv()
.
Use the finish = FALSE
argument to build_table()
to prevent the Markdown
table from being created.!
iris %>% outline_table() %>% add_rows('Sepal.Length', stat_meanSD) %>% add_rows('Sepal.Width', stat_medianIQR) %>% renaming('header', c('Measures', 'Values')) %>% build_table(finish = FALSE) %>% write.csv('table1.csv', row.names = FALSE)
Any scripts or data that you put into this service are public.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.