Improving CS Performance by Developing Spatial Skills

Human information processing and storage. To understand how spatial skills apply to non-spatial reasoning and memory, we must understand how humans reason with and remember information. We have two main systems: working memory and long-term memory.³ Working memory is where humans process new information and reason with information stored in long-term memory. It has an extremely limited capacity.⁴ Long-term memory, by contrast, is where we can store practically endless information, as long as we can figure out how to recall it.

Because working memory has a limited capacity, it has been studied extensively to better understand how to maximize its use. In this research, cognitive scientists found that there are separate systems in working memory for visual and auditory reasoning.³ The visual system includes spatial reasoning and is called the visuospatial sketchpad because it is like a mental sketchpad that requires no verbal representations. Verbal information is processed in the auditory system, called the phonological loop. It therefore makes sense to think of these two systems primarily as verbal and nonverbal. CS and most other STEM fields rely more on nonverbal information and reasoning than non-STEM fields.

While working memory is fundamentally limited, we can develop strategies to use it more effectively. For example, the distinction between novices and experts depends on how much information they can process at once, which is due to the working memory strategies used.⁴ Strategies allow us to rapidly encode new information into long-term memory while allowing us to recall it. The more information we have stored in long-term memory and the better we can recall it, the better we will perform on demanding reasoning tasks. Though these strategies are typically task-specific,²⁸ those learned while training spatial skills have a global effect on nonverbal information and reasoning.⁴⁰

Transferring spatial skills to cognitive skills. Given the robust correlation between spatial skills and STEM performance, spatial training is a common area of research in education. In a review of more than 200 research studies about spatial training, Uttal and colleagues found that spatial training is equally effective for learners of different genders, ages, and levels of initial spatial skill.⁴⁰ Further, the type of training they practiced had no effect on their outcomes. Learners could practice spatial skills by playing a game, taking a course, or practicing spatial tests, and their outcomes were similar. Examining types of training further, a recent study of 17,648 schoolchildren found that practicing spatial encoding (that is, creating mental representations) was more beneficial for mathematics performance than practicing spatial manipulation (that is, visualizing changes to mental representations²²). Though focused on training a single dimension of spatial skills—spatial encoding—this study reified the global benefit of spatial training in any dimension.

Toward explaining this phenomenon, a group of neuroscientists found the physical basis for the application of spatial skills training to advanced nonverbal cognitive skills. This group won a Nobel Prize in 2014 for their discovery that grid and place cells, primitive structures in the brain that allow us to navigate our physical space, are also activated by navigating information that is non-spatial but also nonverbal, resulting in a sort of pseudo-spatial processing.^5,10 Grid and place cells’ functions align with spatial encoding and orientation skills. Grid cells encode and recall mental representations of the world around us. If you are in a familiar room, you could close your eyes and still know where the furniture around you is due to the mental map created by grid cells. In a complementary role, place cells mentally represent where you or another object is in relation to features of the environment. If you are at home, you can probably walk to the bathroom in the middle of the night without turning on the lights due to place cells.¹⁰

A recent theory from CS education, spatial encoding strategy theory, posits that spatial training improves STEM performance by developing strategies to leverage these grid and place cells for pseudo-spatial information and reasoning.²⁵ Pseudo-spatial reasoning includes creating physical representations of non-physical concepts such as personality traits (for example, introverted/extroverted, flexible/rigid, selfish/selfless). Imagining these non-spatial concepts as if they occupied a “place” on a dimension or grid, especially to represent someone’s personality, shows our proclivity for creating pseudo-spatial concepts. We do the same for CS concepts, such as ordering an if-else statement from the most to least restrictive condition.

Parkinson and Cutts theorized that spatial skills are valuable in CS because CS practitioners are frequently required to construct robust, abstract mental representations of structures and processes. They associate spatial visualization skill with the “construction, manipulation, and development of a persistent mental model” and link it and other spatial skills to elements of program comprehension and program generation in CS. These ideas fit closely with the idea that humans recruit spatial structures in the brain for pseudo-spatial representations of complex, abstract ideas.²⁹

The benefit of spatial training, then, is to maximize processing capacity in limited working memory by developing strategies that can be applied to encoding and orienting any pseudo-spatial conceptual representations, such as those used in many STEM fields. In CS, for example, navigating a code base can be equated to navigating a city, and research has found that CS students who are better at navigating a city are also better at navigating a code base¹⁵ and perform better in CS courses.¹⁴ Spatial encoding strategy theory posits that this connection is due to better strategies for processing spatial and pseudo-spatial information. Our grid cells provide mental representations at multiple levels, such as the immediate area in fine detail and the entire area in less detail. Our place cells then provide cues that signal how these two representations relate to each other and where we currently are in those representations. This is just one example of how CS performance is related to spatial performance; the next section will explore this research in detail.

Spatial skills outside computing. Even before computing was generally considered an independent discipline, there were studies associating spatial skills with achievement in computing. In the 1950s, Super and Bachrach reviewed studies in vocational and career research in STEM domains and found spatial visualization and spatial orientation to be recurring predictors of success.³⁸ Although they do not mention computing explicitly—expected given that computing only came to the fore as an independent discipline in the 1960s—Super and Bachrach repeatedly associate spatial skills with other “special aptitudes,” such as “mechanical comprehension” and “manipulative comprehension,” generally considered to be related to CS skills.

Recognizing the connection between spatial skills and STEM performance, researchers began exploring spatial training to improve STEM performance. Beginning in the 1990s, Sheryl Sorby pioneered explicit spatial skills training for engineering students by developing a one-hour weekly course for a 10-week semester.^34,35 The course consisted of visualization software and a workbook, both of which develop spatial skills through sketching and multiple-choice exercises across a range of spatial topics. Sorby’s early offerings were voluntary, but the course improved student outcomes substantially and reliably enough that the course is now mandatory for any Michigan Tech students declaring an engineering major who score below 60% on a spatial skills test. In the past two decades, the course has been applied in several different contexts, and it produces consistently better engineering grades and retention for undergraduate students—particularly women, who historically tend to have lower spatial skills than men in the general population.

It might be expected that spatial skills training would be valuable for a subject that conceptualizes and manipulates physical 3D objects, such as engineering. After all, some of the exercises in the training are very similar to drafting or technical drawing, an essential foundation in engineering graphics. However, Veurink and Sorby have also examined the effects of the spatial training course on students’ performance in courses from other STEM domains.⁴¹ They compared students in the training course against those who scored marginally above the required threshold (60 to 70% on the spatial test) or were not required to because they were not engineering majors. They found that students completing the training course earned significantly higher grades than those who did not in calculus, chemistry, and computer science classes. Therefore, while the training was designed to improve engineering outcomes specifically, it has additional benefits across STEM tasks that are less obviously about manipulating 3D objects and space.

Spatial skills’ correlation with CS outcomes. Since Sorby started working in engineering, several studies have directly associated spatial skills with CS success or aptitude. The typical approach to examining these relationships is to correlate a spatial skills test with CS performance or achievement, typically on programming tasks. Significant, positive correlations have been discovered between mental-rotation (that is, spatial visualization) scores and performance in introductory computing courses in multiple institutions across several years.^6,21,23 This correlation means that students with higher mental-rotation scores also perform better on exams and practical assessments in their introductory CS courses. This correlation holds for students who have little or no programming experience when starting their CS program and those with STEM degrees in other fields who are learning CS for the first time. Spatial skills have also been associated with more-standardized measures than course grades. One study of pre-university CS students found a correlation between mental-rotation scores and scores in a subset of the Advanced Placement Computer Science (AP CS A) test, which can be exchanged for college credit in the U.S.¹¹ Researchers have also found a correlation between spatial skills test scores and the SCS1, a validated introductory CS assessment that tests skills with which any student who completes an intro course should be familiar.⁷

Higher spatial skills also correlate with task-specific measures of computing. Students issued a code-navigation task²⁰ were asked a series of questions about a program that required navigation between 1,000 lines of code spread across six files. Students were instructed to click each line of code as they read it, and the screen was recorded to capture the students’ process. The researchers determined that moving the cursor by more than 10 lines of code qualified as an intra-class jump, while clicking between files was considered an inter-class jump. The count of both kinds of jumps showed a significant positive correlation with spatial skills, indicating that students with higher spatial skills jumped between areas of the code more frequently. Students with higher spatial skills also took less time to complete the exercise. Further, in a study involving introductory CS students,³² an expression-evaluation test was developed consisting of 30 multiple-choice items of increasing difficulty. The test used only simple Python expressions and data structures, which all the students were tested to be familiar with in isolation, and combined them in increasingly complex sequences. The students’ spatial skills scores were positively correlated with scores in the expression-evaluation test, and the correlation was stronger for students with less previous CS experience.

The effect of spatial skills training in CS. Given that spatial skills are correlated with success on so many tasks in CS, the question becomes whether training spatial skills leads to improved outcomes. The evidence suggests that training does improve outcomes in CS undergraduate classes. Two parallel spatial skills training programs at four different institutions both demonstrated potential training benefits.^7,30 Students taking spatial skills training courses alongside their introductory computing coursework performed better than those who did not participate. Even across a short, two-week computing summer school, in which some students were allocated to daily spatial training sessions and others to additional computing sessions, the students undertaking spatial skills training showed higher improvements between the beginning and end of the course.¹¹

Despite reliable improvements, the empirical research on spatial training for CS undergraduates is still limited and does not always demonstrate the same magnitude of effect, probably due to uncontrolled factors. Almost all spatial skills studies in CS have been conducted naturalistically with existing cohorts, which falls short of the research ideal of randomized controlled trials. The strongest correlations between spatial skills and CS have been observed in students in introductory computing at a university level, particularly those with less experience. This aligns with the spatial encoding strategy theory discussed earlier, because it is expected that students gradually develop their own domain-specific strategies and rely less on the general strategies measured by spatial skills tests.

There is some research, however, that appears to indicate a diversion from this specific facet of spatial encoding strategy theory. Parkinson and Cutts found that correlations between spatial skills in students’ first year of study and CS achievement continue all the way up to the final year of students’ graduating degree, and in fact grow stronger as students progress. In one study,³¹ final-year module results in which students encountered completely new conceptual computational theories and models, such as machine learning and functional programming, correlated with spatial skills. In contrast, grades in non-computationally complex modules like human-computer interaction and professional skills and issues did not show significant correlations with spatial skills. In studies of computer graphics modules, where the subject matter of the module is much more closely tied to spatial skills, multiple aspects of spatial skills (particularly visualization and encoding) correlated with module grades. Clearly, the connection between spatial skills and computing achievement is not only an introductory issue but also continues into future study. This finding matches those across STEM education too. Wai, Lubinski, and Benbow’s Project Talent review observed that participants who achieved STEM degrees (up to Ph.D.) were likely to have scored highly on a spatial skills test back in high school: 45% of participants achieving a STEM Ph.D. were in the top 4% of scorers in spatial skills in high school.⁴²

This long-term correlation between spatial skills and CS performance may be due to a knock-on effect of early learning: Students who do not have the generalizable skills to create conceptual representations may in turn struggle to develop domain-specific strategies, affecting their progress further into their degree program. If a student never truly develops an effective, transferable strategy for, to give an extreme example, program comprehension, then any task involving program comprehension throughout their studies will be challenging and time-intensive. This struggle will be further compounded by the assumptions of their peers and instructors, who have developed effective strategies for dealing with program comprehension and will progress through content at speeds and complexities that students with strategy deficits will struggle to match. If this is the case, then it is easy to see how STEM domains, particularly further along in academia, are populated by those with stronger spatial skills. When we also consider that there is evidence of students with higher spatial skills scores performing better in later-stage modules, it is also easy to see why students with less-developed spatial skills may decide to drop out of programs or not pursue more advanced degrees.

Alternatively, it’s possible that domain-specific strategies that students develop may have limitations in progressive CS learning. The strategies learned for introductory programming in a student’s first year may be of limited use in a machine learning or artificial intelligence module in the student’s final year. In this case, students must revert back to their general strategies again, thus depending on the same skills exposed by spatial skills tests. Indeed, perhaps students rely on these general strategies for any new learning, whether it is a new, complex computational model—for example, a completely new programming language paradigm or a new approach to addressing computing problems, such as machine learning—or simply a codebase they have never seen before. The empirical research has little work on understanding how students adjust from using generic skills to domain-specific ones in this context. Therefore, we have little understanding of how far spatial encoding strategy theory goes toward explaining the pervasive spatial skills connection.

Clearly, there is more work to be done in truly understanding how spatial skills and training can be used to improve achievement in computing, but in all the specific studies of spatial skills in CS to date—and even in broader STEM—we observe that spatial skills correlate with success in various measures of computing. Beyond initial learning, we have also seen that spatial skills have long-term benefits for CS students and practitioners, affecting lifelong abilities and skills. This correlation should be interesting to CS students and practitioners because spatial skills can be easily trained, resulting in better skill development. While our next steps in the research may be to unpack this relationship further and identify its constituent parts, it is still clear that focusing on the spatial skills of those learning new CS concepts is a valuable practice.