CODING AS MOTIVATION

Learning statistics can be a difficult process for undergraduate political science students. Many students choose to major in political science, due in part at least to the perceived absence of quantitative requirements for the discipline. Concepts such as null-hypothesis significance testing are far from straightforward and require previous knowledge of probability distributions, which in turn requires knowledge of calculus. Even learning how to read statistical material can be challenging if students are not familiar with mathematical notation. Taken together, this material results in a semester-long journey in which the importance of individual components may not be apparent for many weeks.

In contrast, programming is characterized by immediate feedback and frequent discrete confirmation that students are doing things correctly. Each time a student runs a line of code without encountering an error is a small victory. When students learn to create a scatterplot, they can immediately and intuitively assess the strength of a bivariate linear relationship, whereas computing the correlation coefficient requires first understanding what a z-score is and how to compute it.

Although coding in R is often frustrating, students’ successes during in-class coding sessions can serve as the highlight of instructional time. The tangible nature of the code that students produce in class can also encourage some to apply their new programming skills to projects outside of the course. One student told me that they planned to continue learning R so they could use sports analytics techniques for their fantasy sports team, and another said that they hoped to use R to conduct the analyses for a psychology research project they were planning. The fact that R is open-source software and freely available—unlike less-programming-intensive options such as Stata and SPSS—means that students can easily apply the skills they learn in the course to their own projects. This sense of transferable skills helps students to connect to what can be abstract and disconnected material. Job listings for data science positions mention R twice as frequently as any other statistical software typically taught by political scientists (Li Reference Li2019), and many students are willing to learn it because of this connection to data science.

THE LURE OF DATA SCIENCE

The emergent field of data science represents the intersection of statistics, computer science, and substantive knowledge (Davenport and Patil Reference Davenport and Patil2012). Beyond the basic quantitative skills that employers desire in applicants (National Association of Colleges and Employers 2020), data science skills are in particular demand (Davenport and Patil Reference Davenport and Patil2012). A higher proportion of data scientists have non-computer science backgrounds than many other tech industry jobs (Lindner Reference Lindner2018), making it possible for students without a degree in computer science to pursue these opportunities.

Multiple students in my undergraduate methods course noted in their pre-course surveys that they chose my course to fulfill their methods major requirements instead of less-quantitatively-intense alternatives due to their interest in data science. Many nonmajors also expressed that the possibility of learning data science skills contributed to their decision to take the course. Emphasizing and advertising the programming content of undergraduate quantitative methods courses can be beneficial to departments by boosting enrollment in political science courses.

Students can be discouraged from studying data science because they believe that they must be “very smart” and “good at science and math” to learn computer science (Google 2015). In contrast, political science attracts many students who believe that they are not skilled in math or science. I pursued political science as an undergraduate largely to avoid taking additional math courses, and I did not discover an appreciation for math until my graduate coursework paired it with the motivational benefits of coding and substantive material that interested me. Teaching programming in quantitative methods courses thus represents an opportunity to impart skills—and potentially help students discover a passion for data science or adjacent disciplines—that they otherwise might not acquire. This latter benefit is especially important when we consider which students are least likely to follow a more traditional path to data science.

There is a widespread perception of “coding as an impossibly hard technical skill [that] has been used to push people out of the field and diminish the contributions of whole groups of people” (Shugars Reference Shugars2021). Despite the prevalence of women in the early days of computing, computer science has come to be a largely male-dominated field (Abbate Reference Abbate2012; Beyer Reference Beyer2014). The early contributions of Black engineers and computer scientists to the development of computers and the internet have been similarly marginalized in the popular imagination (McIlwain Reference McIlwain2020). In a study of 7th- to 12th-grade students in the United States, both girls and boys rated boys as considerably more interested in computer science (Google 2015). Male students cite more early exposure to computers as a reason to major in computer science (Funke, Berges, and Hubwieser Reference Funke, Berges and Hubwieser2016) and are more likely to have early exposure to computers (Papastergiou Reference Papastergiou2008). Female students are also less likely to pursue computer science because biased teachers early in their education may steer them toward less-technical fields and weaken their self-confidence (Carlana Reference Carlana2019). In addition, parents who lack experience in computer-related fields may rely on and reinforce societal gender stereotypes (Cheng and Huang Reference Cheng and Huang2016, 281). Black and Hispanic students are more likely to fall off of the advanced math track when transitioning from middle school to high school, even after controlling for academic performance (Irizarry Reference Irizarry2021). Once in high school, they are less likely to be exposed to computer science, and they express lower levels of confidence in their ability to learn computer science (Google 2015). Teaching programming in quantitative political methodology courses represents one way to counteract these forces and to introduce nonmale and nonwhite students to data science who may be deterred from computer science departments.

Due to the surging demand for computer science education in the United States, many departments have responded by enacting competitive enrollment policies (Singer Reference Singer2019). Students in introductory-level computer science courses in these departments cite this competition as one reason for not continuing in the major (Lewis, Yasuhara, and Anderson Reference Lewis, Yasuhara and Anderson2011). Competitive admissions policies also lead to lower levels of self-perceived belonging in the department and self-efficacy for nonwhite and nonmale students (Nguyen and Lewis Reference Nguyen and Lewis2020). Because political science departments rarely limit enrollment in quantitative courses, they are not subject to the same competitive forces. Quantitative methods courses in political science thus represent an alternative pathway for students from historically excluded groups to encounter coding as a practice and data science as a discipline.

INSTRUCTIONAL SETUP

Realizing these benefits in a quantitative methods course requires careful attention to course design to integrate programming skills with substantive material. Simply adding a programming component may lead to worse course outcomes for students who have less exposure to computing due to their gender, race, and socioeconomic-status background. I attempt to accomplish these goals by structuring the course to spend substantial instructional time engaged in collaborative coding activities. Spending more time on programming necessarily entails spending less time elsewhere; therefore, care should be taken that the omitted material is not essential to a basic understanding of research design and statistical inference.Footnote ² This course structure requires at least partially adopting a flipped-classroom model that exposes students to R programming outside of class. Websites such as Dataquest provide a ready-made solution that combines in-depth instruction and the ability to write and test R code in a browser. Another option is the learnr R package, which offers a similar interactive-learning format but requires students to have a working RStudio installation. Effectively using class time to allow students to explore the more complex aspects of coding works best with either a small enrollment or a group of teaching assistants for larger courses. This way, a member of the instructional team is always available to assist with issues as they arise.

Having students work in groups, whether their neighbors for in-class activities or assigned groups for semester-long scaffolded final projects, has many benefits. This type of collaborative programming exercise can be especially helpful for female students, resulting in them feeling more engaged and more confident (Ying et al. Reference Ying, Pezzullo, Ahmed, Crompton, Blanchard and Boyer2019). Working in groups also allows students who quickly grasp a given concept to assist their groupmates, thereby reinforcing their own learning as well as freeing the instructor to assist with more complex issues that other students encounter.

Equally important to assisting students with resolving those issues is demonstrating how I overcame them. When a student has a computing error that I am not able to quickly diagnose and correct, I get the class’s attention and work through finding and implementing the answer. This process typically involves using the classroom projector to show the specific Google search term I use, highlighting the relevant lines in the resulting web page, and then successfully correcting the error in RStudio.

How instructors go about helping students in class is more important than the specific content of any assistance. In pre-course surveys, many students are concerned about the prospect of learning to program. Multiple students noted in course evaluations that a “judgment-free atmosphere” was important to their success in the course. Students often lack fundamental computer-literacy skills (e.g., understanding folder structuresFootnote ³), so an instructor must truly convince them that there are no dumb questions or dumb errors.

One tactic that goes a long way toward accomplishing this latter goal is to convey the frequency and simplicity of one’s own coding errors to students. I often recount ways in which I had failed to correctly use a specific function as students were learning it for the first time. Live-coding through a task on the classroom projector presents ample opportunities for errors to arise organically, allowing the instructor to model how to respond to them. Many students seem paralyzed when encountering error messages; therefore, demonstrating how to respond and resolve them is an invaluable skill that they use when completing problem sets or working on final projects outside of class time. The goal is to use class time as effectively as possible to give students the tools they need to complete work independently.

Because providing this level of detailed attention to students requires time and substantial effort, departments should provide instructional support for those teaching these courses. When departments are unable to offer small class sizes or teaching assistants to instructors, they should provide course releases in recognition of the additional prep work required to successfully teach programming. They also should decrease the weight given to student evaluations for instructors of these courses. Student evaluations are systematically lower for instructors of quantitative courses (Uttl and Smibert Reference Uttl and Smibert2017), and instructors should be not penalized for choosing to teach more demanding courses. Racial and gender biases are endemic to student evaluations (Chávez and Mitchell Reference Chávez and Kristina2020; Mitchell and Martin Reference Mitchell and Martin2018), with one study finding that women experienced the largest difference in evaluations from men when teaching quantitative courses (Mengel, Sauermann, and Zölitz Reference Mengel, Sauermann and Zölitz2018). Departments seeking to support faculty from historically excluded groups should be particularly aware of these dynamics because these faculty face significantly higher demands on their time than faculty from historically included groups (Gewin Reference Gewin2020).