In mark-andrews/psyntur: Helper Tools for Teaching Statistical Data Analysis

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

Introduction

The summarize function in dplyr, especially when combined with group_by and across, provides powerful tools for exploring data using summary statistics. The psyntur package provides some wrappers to these tools to allow data exploration, albeit of a limited kind, to be done quickly and easily. We explore some of these functions in this vignette.

Load the psyntur functions and data sets with the usual library command.

library(psyntur)

Summary statistics with `describe`

We can use the describe function in psyntur. The first argument to describe should be the data frame. Subsequent arguments should be named arguments of summary statistics functions, like mean, median, etc., applied to any variables in the data frame. For example, using the faithfulfaces data frame, we can obtain the arithmetic mean and standard deviation of the faithful variable as follows.

describe(data = faithfulfaces, avg = mean(faithful), stdev = sd(faithful))

We can apply the same or different functions to the same or different variables.

describe(data = faithfulfaces,
         avg_faith = mean(faithful), 
         avg_trust = mean(trustworthy),
         sd_trust = sd(trustworthy))

We can obtain the summary statistics for the chosen variables for each group of a third variable using a by variable.

describe(data = faithfulfaces, by = face_sex, 
         avg = mean(faithful), stdev = sd(faithful))

The by argument may be a vector of variables. In this case, the chosen variables are grouped by the combination of the by variables. For example, in the following we group the time variable in vizverb by both task and response.

describe(vizverb, by = c(task, response),
         avg = mean(time),
         median = median(time),
         iqr = IQR(time),
         stdev = sd(time)
)

Multiple summary functions to multiple variables

It would be tedious and repetitive to use describe as above if wanted to apply the same set of summary statistic functions to a set of variables. Instead, we can use describe_across. For example, to calculate the mean, median, standard deviation to two variables, trustworthy and faithful, in the faithfulfaces data set, we can do the following.

describe_across(faithfulfaces,
                variables = c(trustworthy, faithful),
                functions = list(avg = mean, median = median, stdev = sd)
)

Note that the data frame that is returned is in a wide format. We can pivot this to a longer format by saying pivot = TRUE.

describe_across(faithfulfaces,
                variables = c(trustworthy, faithful),
                functions = list(avg = mean, median = median, stdev = sd),
                pivot = TRUE
)

We can use the by variable to calculate the summary statistics for each subgroup corresponding to each value of the by variable, as in the following example.

describe_across(faithfulfaces,
                variables = c(trustworthy, faithful),
                functions = list(avg = mean, median = median, stdev = sd),
                by = face_sex,
                pivot = TRUE
)

As in the case of describe, the by argument can be a vector of variables.

Dealing with missing values with `_xna`

When variable have NA values, most summary statistics function will, by default, return NA. To illustrate this, we can modify faithfulfaces to contain NA's for the faithful variable.

faithfulfaces_na <- faithfulfaces %>%
  dplyr::mutate(faithful = ifelse(faithful > 6, NA, faithful))

Now, if we try one of the above describe or describe_aross functions with the faithful variable, we will obtain corresponding NA values.

describe_across(faithfulfaces_na,
                variables = c(trustworthy, faithful),
                functions = list(avg = mean, median = median, stdev = sd),
                by = face_sex,
                pivot = TRUE
)

Of course, if we set na.rm = TRUE in any or all of the summary functions, we will remove the NA values before the statistics are calculated. This is relatively easy to do with describe, as in the following example.

describe(data = faithfulfaces, by = face_sex, 
         avg = mean(faithful, na.rm = T), stdev = sd(faithful, na.rm = T))

However, for describe across, we pass in a list of functions, and so to set na.rm = T, we can to create purrr style anonymous functions calling the summary statistic function with na.rm = T, as in the following example.

library(purrr)
describe_across(faithfulfaces_na,
                variables = c(trustworthy, faithful),
                functions = list(avg = ~mean(., na.rm = T), 
                                 median = ~median(., na.rm = T), 
                                 stdev = ~sd(., na.rm = T)),
                by = face_sex,
                pivot = TRUE
)

Anonymous function like this are not very transparent for those new to R, and the resulting function looks quite complex.

In order to avoid using code like ~mean(., na.rm = T), for a number of commonly used summary statistic functions (sum, mean, median, var, sd, IQR), we have made counterparts where na.rm is set to TRUE by default. These functions have the same name as the original with the suffix _xna (but IQR is iqr_xna, not IQR_xna). As such, we can do the following.

describe_across(faithfulfaces_na,
                variables = c(trustworthy, faithful),
                functions = list(avg = mean_xna, median = median_xna, stdev = sd_xna),
                by = face_sex,
                pivot = TRUE
)