What Is a Hive Plot and When Should You Use It for Network Visualization?
A hive plot is a network visualization layout that places nodes on a small number of radial axes and draws edges as curves between those axes. It works best when your network has a few meaningful node categories (roughly 3–9) and you care about the pattern of connections between groups rather than the precise position of every node. If your graph is small, ungrouped, or you need to trace individual paths, a force-directed layout is usually the better choice.
The basic structure
A hive plot trades the free-form scatter of a typical node-link diagram for a fixed geometric frame:
- Axes — straight lines radiating from a common center, one per node category. Each axis is a scale, not a direction.
- Nodes — positioned along an axis according to some value (degree, rank, size, expression level, etc.). Nodes in the same category share an axis.
- Edges — drawn as curved arcs connecting nodes on different axes. Edges within the same axis are usually omitted or handled separately, because they would overlap the axis itself.
Because every node is pinned to a known axis, the layout is deterministic: the same data produces the same picture. That reproducibility is a large part of why the form is used in genomics and other fields where figures need to be comparable across datasets.
The term and the layout were developed by Martin Krzywinski, whose site (mk.bcgsc.ca) documents hive plots alongside related work such as Circos. His genome-comparison pieces, for example "Circles of Life — Genomes Across Time," compare the human genome to those of chimp, dog, opossum, platypus, and chicken — a task where a handful of species become the axes and relationships are read as crossing arcs.
What hive plots are good at showing
The layout is designed to make certain patterns pop out:
| Pattern | How it appears |
|---|---|
| Modularity | Dense bundles of edges between two axes, with little traffic elsewhere |
| Group-to-group flow | Thick bands connecting specific axis pairs |
| Hub nodes | A node with many edges shows as a fan of curves converging on one point |
| Asymmetry | One direction of a relationship is heavy, the reverse is sparse |
| Outliers | A single node far along an axis with a distinctive edge pattern |
If your question is "which groups talk to which, and how much," a hive plot answers it quickly. If your question is "what is the shortest path from A to B," it does not.
Reading one without being misled
Two things cause most misreadings:
Edge crossings are not meaningful. Curves cross constantly by design. A crossing does not indicate a relationship, a conflict, or a hierarchy — it is just geometry. Do not read intersections as data.
Axis order changes the picture. Because edges are drawn between fixed axes, reordering the axes can make the same network look more or less tangled. A clean-looking hive plot may simply have a favorable axis order. When comparing figures, check whether the axes are in the same order; when publishing, state the order and the rule used to assign nodes to axes.
A practical check: if you can't describe the axis assignment in one sentence ("axes are species, ordered by divergence from human"), the figure will be hard for readers to trust.
Hive plot vs. other network layouts
Use the same dimensions to compare:
- Force-directed layouts — good for discovering clusters and for small-to-medium ungrouped graphs; positions are not stable across runs, and large graphs become hairballs.
- Arc diagrams — good for showing edges along a single linear ordering; less suited when nodes fall into several distinct categories.
- Circos-style circular layouts — good for genome-scale data with many tracks and positional relationships; heavier to set up and oriented toward genomic coordinates rather than abstract categories.
- Hive plots — good when categories are few, meaningful, and known in advance, and when you want a stable, comparable figure.
The deciding question is whether your node categories are given by the data. If they are, a hive plot is a strong candidate. If you would have to invent the categories to make the layout work, it will misrepresent the structure.
Deciding whether your data fits
Run through these conditions:
- Do you have 3–9 natural node categories? Fewer than three and the radial structure adds nothing; many more and the axes crowd.
- Is there a sensible value to order nodes along each axis? Degree, weight, time, or rank all work. Without one, node placement is arbitrary.
- Is the interesting signal between categories, not within them? Hive plots handle cross-axis edges well and same-axis edges poorly.
- Do you need a reproducible figure? If yes, the deterministic layout is an advantage over force-directed alternatives.
- Will readers know the axis meaning? If not, add a legend and a stated ordering rule.
If most answers are yes, a hive plot will communicate your network more clearly than a generic node-link diagram. If several are no, start with a force-directed layout and only move to a hive plot once you can name the categories and the ordering value.