Chapter 2 Literature Review

(Myint et al., 2019)

The study sought to determine if visualizations made by beginners are easier to decipher when created in base R or ggplot2. It did so by randomizing Coursera users to complete a study-wide plotting task using either base R or ggplot2, and then had the learners evaluate visualizations crafted by their peers on a few visual characteristics. The users were not completely new to R, but they were still beginners, as they had just completed a course covering basic R usage in both base R and the tidyverse.

The study concluded that the visualizations generated using ggplot2 were often easier to understand when creating a complex multivariate visualization. Students were more likely to uncover insights involving more complex relationships when using ggplot2, where students could use facet_grid() instead of needing the two for() loops for the base R method. Concurrently, more ggplot2 assignees completed the complex visualization task.

This has been the only well-known study conducted that analyzed the differences between the tidyverse and base R for introductory courses. Other arguments have been made on data science blogs favoring base R or the tidyverse when teaching beginning statistics. The following source was written by a proponent of base R.

(Leek, 2016)

Dr. Jeff Leek is a well-established data scientist, biostatistician, and professor at Johns Hopkins Bloomberg School of Public Health. Leek argues for using base R in plotting, and crafts his argument around his three uses for visualizations: exploratory graphs, expository graphs, and grading student work.

Leek believes he has mastered visualizations in base R and can do everything in base R that can be done in ggplot2. For him, ggplot2 does not serve a practical purpose. Leek also cites examples where ggplot() is not compatible with specific visualizations, such as heatmaps, whereas base R can be far more flexible.

Leek also believes that to make production-ready plots, users need to write a large amount of code regardless of the syntax. If anything, the cleanliness of the default ggplot() settings may lead students to believe that the visualizations are suitable for production, when in reality, they would need to add more aspects. By teaching students to create visualizations in base R, students will already be accustomed to writing larger code chunks, and thus will not be phased when needing to write a few extra lines to make the visualization suitable for release.

Leek’s blog post came in response to the writings of some pro-tidyverse instructors, such as Dr. David Robinson, who is an active supporter of using the tidyverse to help teach introductory statistics.

(Robinson, 2014)

Since this post was released prior to the official release of the tidyverse, Robinson, a well-known data scientist and instructor, focuses solely on one of its packages, ggplot2. Robinson argues that ggplot2 should be taught before base R for two reasons:

1. To create equivalent plots in base R that can be easily done using ggplot2, students require an understanding loops and alternative commands. Compared to ggplot(), where students are required to use the aes() function to specify visualization aesthetics, oftentimes, base R plotting requires additional commands. Thus, with limited ggplot2 teaching, students can create interesting visualizations with seemingly advanced attributes that they would not be able to reproduce in base R.

2. ggplot2 forces students to understand tidy data as it is required. Although this may appear to be an unnecessary tool for introductory statistics, Robinson argues that getting students in the habit of forming tidy datasets is good practice for future important functions that also require tidy datasets, such as lm().

After the official release of the tidyverse, Robinson wrote a follow-up blog advocating for the teaching of the entire tidyverse, not just ggplot2, to beginners.

(Robinson, 2017)

Robinson argues for teaching the coding aspect of introductory statistics using the tidyverse syntax based on his past teaching roles. Robinson claims to have been able to teach students with no prior coding knowledge on how to use the tidyverse’s facet_wrap() in 2-3 hours, which is considered a relatively advanced graphical tool.

Robinson believes the role of coding within an introductory statistics course is as follows: to convince students that R is worth learning and to help them bolster their statistical knowledge. With this role in mind, Robinson writes that the tidyverse syntax encourages students to discover insights for provided datasets from the beginning, helping to both convince them of R’s importance and bolster their learning. In contrast, base R coding requires an additional understanding of its general syntax before learning tools to analyze datasets. Eventually, students need to learn base R techniques, but Robinson argues that base R can be taught in combination with the tools provided by the tidyverse, as to not bog down students by general coding rules. For example, functions such as lm() and mean() can be taught alongside the tidyverse’s summarize().

Robinson also argues against the teaching style that prioritizes learning conditionals and loops in these classes, since he believes that students require advanced techniques to make them effective.

Robinson’s blog posts reference tidy dataframes. But what exactly constitutes a tidy dataframe? Dr. Hadley Wickham, long considered to be the tidyverse’s primary creator, explains below.

(Wickham & others, 2014)

Many popular base R functions require tidy data inputs, as well as many tidyverse commands, which were created to work symmetrically with tidy datasets.

A tidy dataset lists each variable as a separate column and every observation as its own row. The rows are ordered by the first variable, measured variables are listed after fixed ones, and related ones are located in adjacent columns to ease comparison.

Wickham describes the benefits of tidy data outside of its requirement for many commands. The framework was inspired by his experience tackling real-world problems and then his subsequent role as a teacher. Wickham argues that most data transformation operations are easiest to perform when each variable is listed as its column. The consistent data structure allows users to start their analysis with the provided dataset instead of spending time working out the logistics of the data. Additionally, there are only a few commands needed to tidy messy data frames, called tidy tools, so once those commands are mastered, it should only take minutes to get directly into the analysis. In Wickham’s words, “Tidy tools are useful because the output of one tool can be used as the input to another.”

Although the above arguments debate reasons for and against using different syntaxes, introductory statistics courses have been generalized by the release and subsequent update of The Guidelines for Assessment and Intruction in Statistics Education (GAISE). The GAISE also provides direction for the programming portion of the course with strict recommendations for how to teach statistical coding.

(Carver et al., 2016)

The GAISE was first endorsed by the American Statistical Association in 2005 and then updated in 2016 to reflect changes in datasets, technology, and professional demand for statistical literacy. The GAISE is recommended for introductory statistics education, but also applies to more advanced statistics courses, as it provides six general recommendations, which already were stated in the 2005 version:

1. Teach statistical thinking
2. Focus on conceptual understanding
3. Integrate real data with a context and a purpose
4. Foster active learning
5. Use technology to explore concepts and analyze data
6. Use assessments to improve and evaluate student learning

The GAISE also states specific goals for students in introductory statistics courses:

1. Students should become critical consumers of statistically-based results reported in popular media, recognizing whether reported results reasonably follow from the study and analysis conducted.
2. Students should be able to recognize questions for which the investigative process in statistics would be useful and should be able to answer questions using the investigative process.
3. Students should be able to produce graphical displays and numerical summaries and interpret what graphs do and do not reveal.
4. Students should recognize and be able to explain the central role of variability in the field of statistics.
5. Students should recognize and be able to explain the central role of randomness in designing studies and drawing conclusions.
6. Students should gain experience with how statistical models, including multivariable models, are used.
7. Students should demonstrate an understanding of, and ability to use, basic ideas of statistical inference, both hypothesis tests and interval estimation, in a variety of settings.
8. Students should be able to interpret and draw conclusions from standard output from statistical software packages.
9. Students should demonstrate an awareness of ethical issues associated with sound statistical practice.

While the GAISE recommends the use of a programming tool as a supplement to the class, it does not state exactly which program students should use. In the next article, prominent professors at Duke University describe how and why they utilize R for statistics courses.

(Çetinkaya-Rundel & Rundel, 2018)

Two instructors at Duke University, Dr. Mine Çetinkaya-Rundel and Dr. Colin Rundel, provide guidelines for how to use statistical programming tools in introductory statistics settings. As the statistical field continues to grow alongside programming tools, programming has morphed into a requirement, especially when handling with the practical and relevant datasets of today. Thus, the question has been shifted from a “should” to a “when” in discussing programming’s role in building a statistical education.

The authors argue that students need to be exposed to programming, preferably R, early in their statistical education, but not on its own. They believe programming should be introduced in a supplemental fashion to give students the opportunity to grasp the topics covered in the traditional learning environment. R helps students bolster their understanding of lectures and also allows for an introduction to programming, as students who choose to progress through future statistics courses will not have to spend time learning basic coding concepts.

In the instructors’ computation classes, they have a goal to computationally visualize something within the first 10 minutes of the first day to draw students to the importance of programming. In terms of software, the authors believe RStudio may help ease the students’ learning curve as an IDE relative to R and due to its auxiliary tools located outside of the main console.

The tidyverse is not the only set of tools crafted to make R easier for beginners. A group of professors created a package designed for introductory programmers, as it equips students with a set of functions that will help them create complex situations with just a few lines of code. The package also provides a connection between base R and tidyverse plots, as it contains functions to switch between base R and ggplot2 visualization settings.

(Pruim, Kaplan, & Horton, 2017)

Building off the tidyverse’s evolution, the mosaic package was created to help beginning coders develop advanced insights, as explained by the package’s authors. The authors believe that mosaic helps students to use programming in R as an asset from the start of their statistical learning. Mosaic depends on a group of packages within the tidyverse suite and contains commands such as mplot(), which allows students to create interactive plots in ggplot2. Mosaic is designed to be implemented alongside the tidyverse.

Similar to how the tidyverse allows for students to start creating in R from the start, mosaic’s authors believe “One of the keys to successfully empowering students to think with data is providing them both a conceptual framework that allows them to know what to look for and how to interpret what they find, and a computational toolbox that allows them to do the looking.” By working through examples immediately, students will begin understand R’s error messages on so they can diagnose their own code issues and lower expectations for perfect code. As a package designed for beginning R users, mosaic is also equipped with specific functions such as rflip(), which helps explain binomial situations without the use of loops.

The final two sources specifically pertain to the analysis conducted in this study. The first is a resource describing the benefits of team-based learning, which has been used in STA 101 courses to help students absorb the material outside of the traditional classroom setting.

(Brame & Director, 2016)

“Team-based learning (TBL) is a structured form of small-group learning that emphasizes student preparation out of class and application of knowledge in class.” Proponents of TBL cite studies concluding that students learned more when placed in TBL settings compared to when they worked individually. TBL may be more beneficial for students because it forces the individual team members to debate amongst each other, thus bolstering or modifying their current knowledge to better comprehend the topic.

Finally, as a check for potential biases, we analyzed the final project assignment documents for each STA 101 section examined in this study.

Source: Final Project Assignment Documents for Each Class

Overall, there were no significant changes in the STA 101 final project assignment document over the course of the 2013-2016 academic years. However, there were still small alterations that may have influenced the project submissions. For final projects completed in 2014 and onwards, student groups were placed in a situation at Paramount Pictures at a hypothetical job, whereas for Fall 2013 students, they were merely assigned the task at hand without the additional setting. Fall 2013 student groups were also not tasked with the audience score prediction component for a movie, but the covariate created in this study was not utilized in the analysis.

Perhaps most importantly, though, the Fall 2013 final project assignment document features a small alteration in the grading section, where Dr. Çetinkaya-Rundel details the factors important in determining the final project grades. In the Fall 2013 semester, one section is titled “Creativity and Critical Thought”, compared to the document for more recent semesters, where the section is titled, “Critical Thought”. Although it is a minute wording discrepancy, the stating of the direct importance of creativity relative to student grades may have provided an extra incentive for Fall 2013 project groups when conducting their EDAs.