Create a Tidy Statistics Output File

Produce a data file containing the output of statistical models and assist with a workflow aimed at writing scientific papers using 'R Markdown'. Supported statistical functions include: t.test(), cor.test(), lm(), glm(), aov(), anova(), and several others. The package is based on tidy data principles and the 'tidyverse' (Wickham, 2017).

tidystats

Authors: Willem Sleegers, Arnoud Plantinga
License: MIT

tidystats is a package to easily create a text file containing the output of statistical models. The goal of this package is to help researchers accompany their manuscript with an organized data file of statistical results in order to greatly improve the reliability of meta-research and to reduce statistical reporting errors.

To make this possible, tidystats relies on tidy data principles to combine the output of statistical analyses such as t-tests, correlations, ANOVAs, and regression analyses.

Besides enabling you to create an organized data file of statistical results, the tidystats package also contains functions to help you report statistics in APA style using R Markdown. Additionally, development has started on a Shiny app and a Google Docs plugin that uses a tidystats data file to report statistics.

Please see below for instructions on how to install and use this package. Do note that the package is currently in development and may contain bugs. If you find any, please let me know by creating an issue here on Github.

tidystats can be installed from CRAN, but the latest version can be installed from Github using devtools.

library(devtools)
install_github("willemsleegers/tidystats")

Setup

Load the package and start by creating an empty list to store the results of statistical models in.

library(tidystats)
 
results <- list()

Usage

The main function is add_stats(). The function has 2 necessary arguments:

• output: The output of a statistical test you want to add to the list (e.g., the output of t.test() or lm())
• results: The list you want to add the statistical output to

Optionally you can also add an identifier, type, a subset of the statistics, whether the analysis was confirmatory or exploratory, and additional notes using the identifier, type, statistics, confirmatory, and notes arguments, respectively.

The identifier is used to identify the model (e.g., 'weight_height_correlation'). If you do not provide one, one is automatically created for you.

The type argument is used to indicate whether the statistical test is a hypothesis test, manipulation check, contrast analysis, or other kind of analysis such as descriptives. This can be used to distinguish the vital statistical tests from those less relevant.

The statistics argument is used to select a subset of statistics that you want to add to the results list, in case this is desired.

The confirmatory argument is used to indicate whether the test was confirmatory or exploratory. It can also be ommitted.

The notes argument is used to add additional information which you may find fruitful. Some statistical tests have default notes output (e.g., t-tests), which will be overwritten when a notes argument is supplied to the add_stats() function.

Supported statistical functions

Package: stats

• t.test()
• cor.test()
• lm()
• aov()
• chisq.test()
• wilcox.test()

Example

In the following example we perform several statistical tests on a data set, add the output of these results to a list, and save the results to a file.

The data set is called cox and contains the data of a replication attempt of C.R. Cox, J. Arndt, T. Pyszczynski, J. Greenberg, A. Abdollahi, and S. Solomon (2008, JPSP, 94(4), Exp. 6) by Wissink et al. The replication study was part of the Reproducibility Project (see https://osf.io/ezcuj/). The data set is part of the tidystats package.

# Paired t-test
M1_condition <- t.test(call_parent ~ condition, data = cox, paired = TRUE)
results <- add_stats(M1_condition, results)
 
# Correlation
M2_parent_siblings <- cor.test(cox$call_parent, cox$call_siblings, alternative = "greater")
results <- add_stats(M2_parent_siblings, results)
 
# Regression
M3_condition_anxiety <- lm(call_parent ~ condition * anxiety , data = cox)
results <- add_stats(M3_condition_anxiety, results)
 
# ANOVA
M4_condition_sex <- aov(call_parent ~ condition * sex, data = cox)
results <- add_stats(M4_condition_sex, results)

Having added the statistical results to the list, you can convert the list to a table or to a data file, ready for sharing. The example below shows how to produce a table containing all of the statistical results.

library(dplyr)
library(knitr)
options(knitr.kable.NA = '-')
 
results %>%
  stats_list_to_df() %>%
  select(-notes) %>%
  kable()

To write the results to a file, use write_stats() with the results list as the first argument.

write_stats(results, "data/results.csv")

This produces a .csv file that can be shared and that can also be used to write your Results section. The report functions will be demonstrated below.

Report functions

To start reporting your results, first load in the previously saved data file containing the results. This will create a list, just like it was when it was originally saved.

results <- read_stats("data/results.csv")

Additionally, you can use options() to set the default tidystats list to use. This way the report() functions below require one fewer argument. You set the default tidystats list by running the following code.

options(tidystats_list = results)

The main function for reporting is report(). To figure out how to report the output in APA style, tidystats uses the method information stored in the results list. For example, the model with identifier 'M1' is a paired t-test. tidystats will parse this, see that it is part of the t-test family, and produce results accordingly. tidystats() also has test-specific reporting functions, such as report_t_test() that are used under the hood, but they are also available for you to use.

Below we show a list of common report examples:

As you can see in the examples above, you can use report() to produce a full line of output when a model identifier is provided (and a term when the model consists of multiple terms). You can also only retrieve a single statistic by using the statistic argument. Additionally, you can refer to terms using either the term label or the term number. Although this latter method might be less descriptive, it reduces the amount of code clutter in your Markdown document. Our philosophy is, in line with Markdown's general writing philosophy, that the code should not distract from writing. To illustrate, writing part of a results section will now, using tidystats look like this:

To execute the code, the code segment should be surrounded by backward ticks (see http://rmarkdown.rstudio.com/lesson-4.html), which results in:

We found no significant difference between the mortality salience condition and the dental pain condition on the number of minutes allocated to calling one's parents, t(99) = -1.26, p = .21.

Helper functions

Descriptives

Since it's common to also report descriptives in addition to the statistical results, we have added a hopefully useful describe_data() and count_data() function to calculate common descriptive statistics that can be tidied and added to a results data frame. Several examples follow using the cox data.

# Descriptives of the 'anxiety' variable
describe_data(cox, anxiety)

## # A tibble: 1 x 13
##   var     missing     n     M    SD     SE   min   max range median  mode
##   <chr>     <int> <int> <dbl> <dbl>  <dbl> <dbl> <dbl> <dbl>  <dbl> <dbl>
## 1 anxiety       0   200  3.22 0.492 0.0348  1.38  4.38     3   3.25   3.5
## # ... with 2 more variables: skew <dbl>, kurtosis <dbl>

# By condition
cox %>%
  group_by(condition) %>%
  describe_data(anxiety)

## # A tibble: 2 x 14
## # Groups:   condition [2]
##   var     condition     missing     n     M    SD     SE   min   max range
##   <chr>   <chr>           <int> <int> <dbl> <dbl>  <dbl> <dbl> <dbl> <dbl>
## 1 anxiety dental pain         0   100  3.26 0.497 0.0497  1.62  4.38  2.75
## 2 anxiety mortality sa…       0   100  3.17 0.485 0.0485  1.38  4.38  3   
## # ... with 4 more variables: median <dbl>, mode <dbl>, skew <dbl>,
## #   kurtosis <dbl>

# Descriptives of a non-numeric variable
count_data(cox, condition)

## # A tibble: 2 x 4
##   var       group                  n   pct
##   <chr>     <chr>              <int> <dbl>
## 1 condition dental pain          100    50
## 2 condition mortality salience   100    50

If you use the describe_data() and count_data() function from the tidystats package to get the descriptives, you can use the tidy_describe_data() and tidy_count_data() function to tidy the output, and consequently add it to a results list.

results <- cox %>%
  describe_data(anxiety) %>%
  tidy_describe_data() %>%
  add_stats(results, identifier = "anxiety", type = "d", notes = "Anxious attachment style")

In the add_stats() function you can also specify which of the statistics you would like to store in the results list, using the statistics argument. Of course, the results can also be tidied when the data is grouped.