From Code to Concept: Evaluating Multiple Coordinated Views in Introductory Programming

TL;DR

This study investigates how novice programmers in a large, diverse introductory Python course form abstract mental models and cope with cognitive load. It compares a Multiple Coordinated Views (MCV) visualization—synchronizing code, memory, and conceptual analogies—with a Single Visual (SV) tool and Text-Only (TO) explanations over a 12-week deployment involving $N = 829$ students. Results show that while text-only conditions incur higher immediate mental effort, overall cognitive load differences are minimal, and the MCV approach yields consistently higher engagement, especially for learners facing language barriers or with lower prior experience. The findings emphasize adaptive, interconnected external representations that balance concrete fidelity and abstract reasoning, offering design heuristics for scalable, inclusive visualization tools in programming education.

Abstract

Novice programmers often struggle to understand how code executes and to form the abstract mental models necessary for effective problem-solving, challenges that are amplified in large, diverse introductory courses where students' backgrounds, language proficiencies, and prior experiences vary widely. This study examines whether interactive, multi-representational visualizations, combining synchronized code views, memory diagrams, and conceptual analogies, can help manage cognitive load and foster engagement more effectively than single-visual or text-only approaches. Over a 12-week deployment in a high-enrolment introductory Python course (N = 829), students who relied solely on text-based explanations reported significantly higher immediate mental effort than those using visual aids, although overall cognitive load did not differ significantly among conditions. The multi-representational approach consistently yielded higher engagement than both single-visual and text-only methods. Usage logs indicated that learners' interaction patterns varied with topic complexity, and predictive modelling suggested that early experiences of high cognitive load were associated with lower longer-term perceptions of clarity and helpfulness. Individual differences, including language proficiency and prior programming experience, moderated these patterns. By integrating multiple external representations with scaffolded support adapted to diverse learner profiles, our findings highlight design considerations for creating visualization tools that more effectively support novices learning to program.

Figures (13)

• Figure 1: Left: The Multiple Coordinated Views (MCV) tool integrating synchronized code, memory diagrams, and abstract analogies in a step-by-step trace. Right: Python Tutor (Single Visual condition), which provides a step-by-step execution trace with memory visualization.
• Figure 2: Course timeline: Weeks 1--8 show key topics; blue dots mark survey points; red dots mark assessments (Term Test 1 in Week 7, Term Test 2 in Week 11, Final Exam after Week 12).
• Figure 3: Differences in final cognitive load and engagement outcomes by representation type. Notes: The x‐axis denotes participants’ self‐reported mental effort on a 7‐point scale (1 = very low effort, 7 = very high effort). The y‐axis (violin width) indicates the distribution density of responses. We also overlay engagement scores, measured on a 5‐point scale (1 = low, 5 = high).
• Figure 4: Effects of initial cognitive load on final mental effort, enjoyment, and visualization clarity. Raincloud plots display distributions, quartiles, and individual responses across Low, Medium, and High initial cognitive load groups.
• Figure 5: K-means clustering of user event frequencies for sorting tasks. The elbow method confirmed that three clusters provided the optimal balance of variance explained and model parsimony.