Writing User-Defined Summary Functions

Overview

The package tangram.pipe can be used to iteratively build a table which allows each row to be uniquely customized. All the possible changes can be seen in the package’s main vignette, “Customizeable Table Building with Tangram Pipe”. One main package feature which is not discussed there is that a user may write their own summary function for the table rows. By default, the package will use default summary functions to calculate a 5-number summary, plus the mean and standard deviation, for numeric data; column-wise proportions are generated for categorical and binary rows. Currently, there are a total of four prewritten numeric summary functions, as well as three prewritten functions for both categorical and binary data. However, it is often the case that a user wants to have increased flexibility and format tangram.pipe output tables in a different way than provided by the currently-available options. This document is intended to walk a user through how to write a custom-made summary function, as well as some suggested inputs and outputs to include for user-defined summary functions using tangram.pipe.

Function Inputs

To see how the default functions for summarizing data work, let’s take a look at the function usage for summarizing numerical data, num_default.

num_default(dt, rowlabel, missing, digits)

All prewritten summary functions for numerical and categorical data take on a generic form such as num_default(dt, ...), where only the argument dt is required. However, in order for these summary functions to work correctly, a total of four arguments are passed to each of the functions.

  1. dt: The dataset to use for the function must be passed into the summary tool. However, the full dataset cannot be implemented into the summary function. Based on how the summary is used in the row functions, the first column of dt must contain the row information for the table; the second column should include the table’s column information, if applicable. Be sure that dt is a dataframe object.

  2. rowlabel: This is the label you want to use for the row in the table. It should match the rowlabel you specify in the row-defining function.

  3. missing: A binary TRUE/FALSE variable which tells the function whether or not to account for missing data. It should match the designation for missing data from the row function.

  4. digits: The number of significant digits to use in the summary.

Each of the prewritten functions use the ellipsis (...) in place of the final three arguments to provide flexibility in writing custom functions. To write your own function, the bare minimum requirement is that you provide an argument for the dataframe object to use in the summary. The remaining arguments rowlabel, missing, and digits are highly recommended to use within your custom function because it is within the summary that these values, which are specified in the row-initialization, are implemented. If you do not include them in your summary function, your specifications for these arguments in the row initialization will not be present in your table for the row of interest. Therefore, it is highly recommended, but not required, for you to include these inputs in your summaries; excluding them will not break the package.

When you write a custom function, be sure to include all arguments outside dt within the ellipsis (...). This is because tangram.pipe’s for functions will input values for rowlabel, missing, and digits as done in the prewritten functions. To provide additional flexibility, using (...) as the second argument following dt will allow for differing arguments to be used while preventing the custom function from inadvertently breaking the package row functions. You can call your inputs within the function body by inputting (...) into a list and calling the elements of (...).

Note that none of the summary function’s arguments should have default values. Since summary functions are called within the row functions from tangram.pipe, the function will end up taking the values already entered into the row function, so be sure not to use default values here. Below is an example usage of a generic summary function, summary_generic using the iris dataset. Here, we want Sepal.Length to be the row variable and Species to be the columns of the main table. Suppose we want to call the row variable “Sepal Length (cm)”, account for missing data, and use 2 significant digits. First, we show the format the data (dt) must be in to pass it to the function

iris %>% 
  select(Sepal.Length, Species) %>%
  head() %>%
  kable(escape = F, align = 'c') %>%
  trimws() %>%
  kable_styling(c("striped", "bordered"))
Sepal.Length Species
5.1 setosa
4.9 setosa
4.7 setosa
4.6 setosa
5.0 setosa
5.4 setosa

Note that the row variable is on the left and the column variable is to the right. If you wanted to avoid splitting by Species, you would only pass the Sepal.Length information into the summary function.

Now, we show the code input needed for our generic summary function.

summary_generic(dt = iris %>% select(Sepal.Length, Species), 
                rowlabel = "Sepal Length (cm)", 
                missing = TRUE, 
                digits = 2)

Body of the function

When writing your summary function, it is important to take note of a few aspects the function should be sure to incorporate within its text. The first important check to make is whether your data includes a column variable or not. The data will be structured differently depending on whether or not your data has two columns or only one, so be sure your function can handle both types of data.

Second, you will likely want to label your variable using the name specified in rowlabel. It is in the summary that the rowlabel specified in the row-initializing function is added to the table, so if you neclect this step, the final table will not have the label the user specifies during row initialization.

Next, the function needs to calculate the amount of missing data if missing = TRUE in the function. As with the rowlabel function, missing data handling is specified in the row initialization but calculated within the summary function, so be sure to write the function in such a way that missing data can be handled if specified.

Finally, be sure that all summary statistics are rounded based on the digits argument. The round and sprintf functions are common tools used to accomplish this so the table output can have a polished look.

Required output

At a minimum, the output of each function should be a dataframe object. Any other object type will cause the row function to fail since the final table, as well as any comparison tests, need dataframes to combine the results together into the finished product. The rightmost column should also be the “Overall” column which contains the summary statistics for the dataset as a whole without accounting for the table’s column variable. This is because the row functions will eliminate this column if the user sets overall = FALSE during row initialization.

So long as the above two requirements are met, the summary function will not break the preexisting functions of tangram.pipe. However, there are certain naming recommendations so that the table is formatted well. Ideally, the row name should be in the first column, called “Variable”. A column labelling the type of measurement, called “Measurement”, as well as naming the overall column “Overall”, will keep the column names consistent with the package’s default summary functions. If you decide to use different names, it is recommended that you keep the names consistent with each new summary function that you use and that you do not mix in rows with default summary functions; mixing in different naming conventions will cause information that should be contained in one column being spread out over multiple columns.

Below are example outputs from the preexisting num_default and cat_default summary functions. It is recommended that you include columns for the variable name, the measure type, the column categories (if applicable), and the overall column, being sure to keep naming conventions consistent. As of version 1.1.0, tangram.pipe default summaries now also calculate the total number of instances, N.

iris %>%
  select(Sepal.Length, Species) %>%
  num_default(rowlabel = "Sepal Length (cm)", missing = TRUE, digits = 2) %>%
  kable(escape = F, align = 'l') %>%
  trimws() %>%
  kable_styling(c("striped", "bordered"))
Variable Measure setosa versicolor virginica Overall
Sepal Length (cm) min 4.30 4.90 4.90 4.30
Q1 4.80 5.60 6.30 5.10
median 5.00 5.90 6.50 5.80
Q3 5.20 6.30 6.95 6.40
max 5.80 7.00 7.90 7.90
mean 5.01 5.94 6.61 5.84
SD 0.35 0.52 0.64 0.83
Missing 0 0 1 1
iris %>%
  mutate(Stem.Size = sample(c("Small", "Medium", "Medium", "Large"), size=150, replace=TRUE)) %>%
  select(Stem.Size, Species) %>%
  cat_default(rowlabel = "Stem Size", missing = TRUE, digits = 2) %>%
  kable(escape = F, align = 'l') %>%
  trimws() %>%
  kable_styling(c("striped", "bordered"))
Variable Measure setosa versicolor virginica Overall
Stem Size Col. Prop. (N)
Large 0.22 (11) 0.24 (12) 0.18 (9) 0.21 (32)
Medium 0.52 (26) 0.44 (22) 0.49 (24) 0.48 (72)
Small 0.26 (13) 0.32 (16) 0.33 (16) 0.30 (45)

Special note regarding binary rows

Binary row summary functions differ slightly from numerical and categorical rows because tangram.pipe’s prewritten summary functions include two additional arguments.

binary_default(dt, reference, rowlabel, compact, missing, digits)

For binary rows, it is recommended that you include the following two argument as well when writing your own functions:

  1. reference: This is the reference variable to include in the table. Since binary data only includes two possible categories, the row function is written so that only one option will be included in the table. The category you want in the table is the value of reference.

  2. compact: Often, binary data in tables is written so that the variable name is eliminated and only the reference group appears in the table. This compacts the row information into a single for. This TRUE/FALSE variable decides if this is how you want the data displayed in the table.

Both of the above variables should be included in the body of the user-defined function so that each is dealt with accordingly. As with numerical and categorical data, you are not required to account for these arguments in the body of your function, but excluding them will result in the reference and compact arguments defined in binary_row to not be implemented in your table output object.

Since writing functions for binary data can be somewhat more complicated, remember that any data used in a binary row can also be substituted into a categorical defined row instead.

User-Defined Comparison Tests

A similar process can be used to write custom functions for comparison tests in tangram.pipe. The user is encouraged to look up the help documentation to prewritten tests for their desired row to determine what arguments are necessary for a custom function to include as input.