Do Boxes Affect Exploration Behavior and Performance in Group-in-a-box Layouts?

TL;DR

The paper investigates how GIB visualization elements—specifically box size and the presence of boxes—affect task accuracy and exploration using an eye-tracking experiment that controls internal-edge density. By systematically varying side-ratio and density congruence across seven groups and multiple stimuli, the study shows that box size can positively or negatively impact accuracy depending on density cues, while the presence of boxes does not uniformly alter exploration behavior. Eye-tracking reveals that participants tend to focus on the internal edges rather than the boxes themselves, indicating a congruency-based effect in numerosity judgments. The findings offer empirical guidance for GIB design, including potential alternatives like heatmap-encoded edge information to reduce box-size confounds and improve interpretability in practical visualization tasks.

Abstract

The group-in-a-box (GIB) layout is an efficient graph drawing method designed to visualize the group structure of graphs. The layout communicates group sizes and both within-group and between-group network structures simultaneously. The layout is characterized by its composition of multiple elements, including nodes, edges, and boxes. However, there is limited empirical guidance on how these elements should be combined. In this paper, we measured participants' task performance and eye movements while identifying the group with the largest number of internal edges. We investigated the effect of visualization elements on task performance while controlling the density of internal edges and the box size. The results revealed that the box size in a GIB layout significantly affects the task accuracy either positively or negatively while eye-tracking data suggests that participants focused on internal edges, not the box size. These findings contribute empirical guidance for GIB layout design and lay the groundwork for future research as GIB layout becomes more widely used.

Figures (9)

• Figure 1: Example of GIB layout. If X (formerly Twitter) data is visualized with the GIB layout, nodes represent users, edges represent messages exchanged between users, and boxes represent communities formed by users who communicate frequently.
• Figure 2: Examples of each task. (a) Task 1: Identify the group with the largest number of nodes when only nodes and boxes are displayed. (b-1) Task 2 (without-boxes): Identify the group with the largest number of internal edges when only nodes and edges are displayed. (b-2) Task 2 (with-boxes): Identify the group with the largest number of internal edges when the original GIB layout is displayed.
• Figure 3: Example of the side ratio ($SR$) for the correct answer candidates. These boxes share the same long side, while their shorter sides are proportional to the number of nodes.
• Figure 4: Explanation of the Density Difference ($\Delta D$). The box on the left contains $N_1$ nodes, an area of $C_1$ for the circle, and $E_1$ internal edges. The box on the right contains $N_2$ nodes, an area of $C_2$ for the circle, and $E_2$ internal edges. The density of internal edges ($D$) is defined as $D_1$ for the left box and $D_2$ for the right box. The density difference ($\Delta D$) is calculated as $D_1 - D_2$. In this example, the left box has a higher density of internal edges than the right box.
• Figure 5: Examples of the density difference between the groups with $E_1$ and $E_2$ edges when the side ratio is 0.91. The group with more internal edges is marked with a gray border. When $\Delta D$ is less than 0, the group with the second-largest number of nodes ($N_2$) has more internal edges ($E_2$), whereas when $\Delta D$ is greater than 0, the group with the largest number of nodes ($N_1$) has more internal edges ($E_1$).