GroupThink

Table of Contents

• Introduction and Install
  • Dependencies
• Benefits of GroupThink
• Core Functionality
  • The unify() Function
  • Left Out Responses
  • Data for unify()
  • View Column Index Numbers
  • The assess() Function
• Further Functionality
  • Aggregate Across Multiple Columns/Questions
  • Make Formatted Tables
  • Filter Out Responses From the Output Only
  • Integrate with ggplot
• Even More Functionality
• Future Plans
• Bug Reports and Feature Requests

Introduction and Install

GroupThink is a package designed to assist in the analysis in categorical survey data. It mainly acts as an interface for existing tidyverse functions - but makes it easier to aggregate responses, do cross-question analysis, and avoid classic mistakes typical of survey data analysis.

It currently has two functions:unify() and assess() (though others are planned for the future…!).

GroupThink isn’t on CRAN, so you’ll need to use devtools to install it. Run:

install.packages("devtools")
library(devtools)

devtools::install_github("Samuel-Osian-Andrews/GroupThink")
library(GroupThink)

As GroupThink is still in development, you should periodically reinstall the package to get updates.

 

Dependencies

GroupThink depends on dplyr, tidyr and gt libraries. If these aren’t installed automatically when you install GroupThink, you may need to run:

install.packages(c("dplyr", "tidyr", "gt"))

 

Benefits of GroupThink

GroupThink is a response to key bottlenecks and common mistakes when analysing survey data. The function is beneficial because it…

• Allows for easy groupings. unify() makes it very easy to group together different Likert-style responses (e.g. combining Somewhat agree with Strongly agree, or Excellent and Very good). It’s now extremely difficult to make mistakes with incorrect groupings, as unify() alerts you of any unassigned responses.

• Automates your calculations. unify() handles n and proportion calculations for you, meaning you no longer have to undertake complex data manipulation tasks, avoiding functions such as pivot_longer(), which can result in inaccurate figures if you’re not careful!

• Works with full questions as column headers. Typically, exporting survey responses (such as from Microsoft Forms or SurveyMonkey) will leave you with full questions (e.g. “Do you agree or disagree that…”) as column headers. This is usually a nightmare to work with in R. Because unify() works on column indexes, rather than column names, you don’t need to worry about recoding your columns or typing out full survey questions throughout your code.

• Gives usable outputs.. unify() neatly integrates with ggplot, allowing you to visualise your aggregated data. Alternatively, you can produce formatted tables through the gtTable argument.

• Presents clear, readable syntax. Even for those unfamiliar with R syntax, unify() makes it very clear exactly how you’ve grouped together your responses, improving readability and reproducibility.

• Means faster insights. With just a few lines of code, this function could save you hours worth of work for large survey projects.

 

Core functionality

The unify() function

The unify() function groups together Likert-style responses for a given question or set of questions, returning a summarised output that contains the n and proportion for each of these groupings.

unify(data, cols = 1, # ...dataframe name and column index number(s) to analyse
      
      # Below, we 'group' responses via custom grouping labels (e.g. 'Agree'):
      Agree = c("Somewhat agree", "Strongly agree"),
      Disagree = c("Somewhat disagree", "Strongly disagree"),
      Neutral = "Neither agree nor disagree",

      ignore = "Don't know") # ...optionally, set response(s) to ignore from calcs

## # A tibble: 1 × 7
##   Question         `Agree (n)` `Disagree (n)` `Neutral (n)` `Agree (Proportion)`
##   <chr>                  <int>          <int>         <int>                <dbl>
## 1 I find the cour…          38             32            18                 43.2
## # ℹ 2 more variables: `Disagree (Proportion)` <dbl>,
## #   `Neutral (Proportion)` <dbl>

The grouping labels can be anything you like. For example, Agree could instead be Positive, Good, Satisifed or something else entirely. Similarly, Don't know could be its own group, instead of being ignored. You may include as many grouping labels as you’d like.

There’s of course nothing wrong with having just 1 response option per group (e.g. "Somewhat agree" = "Somewhat agree"). The main purpose of unify() is that it forces you to be intentional with how you handle your data, to improve consistency and avoid mistakes.
 

Left out responses

If you forgot to include a response in your custom groupings, unify() will throw an error. This is crucial for avoiding mistakes in your proportion calculations. For example:

unify(data, 1, Agree = "Somewhat agree",
                #"Strongly agree"), -- let's stop unify() from seeing this line
      Disagree = c("Somewhat disagree", "Strongly disagree"),
      Neutral = "Neither agree nor disagree",
      ignore = "Don't know") 

## Error in unify(data, 1, Agree = "Somewhat agree", Disagree = c("Somewhat disagree", : The following responses are not accounted for in the provided groups: 'Strongly agree'. Please check your grouping arguments.

As seen above, the output tells you that you forgot to assign “Strongly agree” to a grouping variable.

These errors are crucial, since other R functions do not warn you if you haven’t accounted for a group, or mistyped “Strongly Agree”, as “Strongly agree”, for example.
 

Data for unify()

The unify() function expects data that looks like this:

## # A tibble: 10 × 3
##    I find the course material en…¹ The course workload …² Feedback from assign…³
##    <fct>                           <fct>                  <fct>                 
##  1 Somewhat disagree               Agree                  Disagree              
##  2 Strongly agree                  Disagree               No opinion            
##  3 Somewhat agree                  Agree                  <NA>                  
##  4 Somewhat disagree               Unsure                 Disagree              
##  5 Somewhat agree                  <NA>                   Strongly agree        
##  6 Strongly agree                  Neither agree nor dis… <NA>                  
##  7 Strongly disagree               Neither agree nor dis… Strongly agree        
##  8 Somewhat agree                  Highly disagree        Strongly agree        
##  9 Somewhat disagree               Neither agree nor dis… Strongly disagree     
## 10 Somewhat agree                  Neither agree nor dis… Agree                 
## # ℹ abbreviated names: ¹​`I find the course material engaging and relevant.`,
## #   ²​`The course workload is manageable within my schedule.`,
## #   ³​`Feedback from assignments is helpful for my learning.`

Responses do not need to be consistently labelled either within or between different questions/columns, and can contain missing data (you’ll likely want to assign NA to the ignore parameter).
 

View column index numbers

Since GroupThink functions work with column index numbers, not column names, you’ll likely want to summarise all index numbers of your dataset. For this, run colnames() from base-R.

colnames(data)

## [1] "I find the course material engaging and relevant."    
## [2] "The course workload is manageable within my schedule."
## [3] "Feedback from assignments is helpful for my learning."

 

The assess() function

You might find it beneficial to run GroupThink’s assess() function, which provides an overview of the different response options in your specified columns.

assess(data, cols = c(2, 3))

## Examining the following columns:
## 
## Column [2]: The course workload is manageable within my schedule.
## Column [3]: Feedback from assignments is helpful for my learning.
## 
## 
## 11 unique responses were found across these columns. Please make sure each response
## below is accounted for within your `unify()` function call:
## 
## 
## Agree
## Disagree
## Highly agree
## Highly disagree
## Indifferent
## Neither agree nor disagree
## No opinion
## Strongly agree
## Strongly disagree
## Unsure
## NA

 

Further functionality

Aggregate across multiple columns/questions

You are not restricted to analysing just one question/column with unify(). You can specify multiple columns/questions to use for the output:

unify(data, c(1, 2, 3), # ...analyse Columns 1, 2 and 3
      Positive = c("Somewhat agree", "Strongly agree", "Highly agree", "Agree"),
      Negative = c("Somewhat disagree", "Strongly disagree", "Highly disagree",
                   "Disagree"),
      ignore = c(NA, "Don't know", "Unsure", "Neither agree nor disagree",
                 "No opinion", "Indifferent"),
      
      hideN = TRUE) # ...(optional) hide n column from output (a lot cleaner!)

## # A tibble: 3 × 3
##   Question                         Positive (Proportion…¹ Negative (Proportion…²
##   <chr>                                             <dbl>                  <dbl>
## 1 Feedback from assignments is he…                   57.8                   42.2
## 2 I find the course material enga…                   54.3                   45.7
## 3 The course workload is manageab…                   36.7                   63.3
## # ℹ abbreviated names: ¹​`Positive (Proportion)`, ²​`Negative (Proportion)`

…Just make sure that you’ve accounted for each response option across your range of columns, otherwise you’ll get an error.

 

Make formatted tables

Using the gtTable argument, unify() makes it simple to create nice, formatted tables.

unify(data, 1, Agree = c("Somewhat agree", "Strongly agree"),
      Disagree = c("Somewhat disagree", "Strongly disagree"),
      Neutral = "Neither agree nor disagree",
      ignore = "Don't know",
      filter = c("Agree", "Disagree"),
      hideN = TRUE, # ...optionally, hide N column
      
      gtTable = TRUE) # ...set gtTable to TRUE

Question	Agree (Proportion)	Disagree (Proportion)
I find the course material engaging and relevant.	43.2	36.4

 

Filter out responses from the output only

If you want to only display one response option in the output, we can use the filter argument.

Note that this is different from the ignore parameter: filter removes unwanted responses after the calculations have been performed, while ignore removes them before.

unify(data, 3,
      Agree = c("Agree", "Strongly agree"),
      Disagree = c("Disagree", "Strongly disagree"),
      Neither = c("No opinion", "Indifferent"),
      ignore = c(NA, "Don't know"),
      
      filter = "Agree") # ...only include the Agree group in the output

## # A tibble: 1 × 3
##   Question                                      `Agree (n)` `Agree (Proportion)`
##   <chr>                                               <int>                <dbl>
## 1 Feedback from assignments is helpful for my …          37                 38.9

The other variable groupings are used for the calculations, but only “Agree” responses are shown in the final output.

 

Integrate with ggplot

Unless you’ve set unify()’s gtTable() argument to TRUE, it will output as a tibble. This means it integrates neatly into ggplot() function calls.

Let’s pretend we’ve already run unify() on columns 1, 2 & 3, and assigned it to the name united…

ggplot(data = united, # ...unify() output becomes ggplot()'s data argument
       aes(x = Question, y = `Agree (Proportion)`, fill = Question)) +
  geom_col() +
  
  # ...below are just optional customisation options:
  coord_flip() +
  theme_bw() +
  scale_fill_manual(values = c("cornflowerblue", "coral", "chartreuse3")) +
  scale_y_continuous(limits = c(0, 70)) +
  theme(legend.position = "none")

 

Even more functionality

For other functionality not covered in this document, please run ?unify() and ?assess() to view the help files, which covers all function parameters.

 

Future plans

• Add unify() support for demographic comparison (comparing figures between factors such as age, sex, etc.)

• Add support for stargazer tables into unify().

• Develop a separate function for analysing multiple choice data for data formats typical of exported survey data.

 

Bug reports and feature requests

Please do let me know of any issues you come across. You can use the Issues tab in GitHub for any bug reports.

If you have any ideas for existing features, or perhaps even new ones, then I’d love to hear them. Let me know in the Discussions tab in GitHub.