How visualisation works
This page explains the calculations behind Visualising a map. You don’t need it to use the tool.
How hard each connection pulls
Section titled “How hard each connection pulls”The layout treats each pair of connected nodes as joined by a spring. Draw the map from decides which connections count and how strong each pair’s spring is.
| Setting | A connection counts if… | A pair’s strength |
|---|---|---|
| All connection types, Count once | it has a type or a note, and its type isn’t left out | Always the same |
| All connection types, Pull closer the more they have | as above | Its number of connections, compared with the busiest pair on the map |
| One connection type, Connected or not | it has that type | Always the same |
| One connection type, Weighted by a value | it has that type | The total of its values (such as total grants), compared with the largest total on the map |
| Custom mix | its type has a score above 0 | See below |
Connections that don’t count are hidden and don’t pull at all.
Custom mix
Section titled “Custom mix”Each type contributes its score once per pair, however many connections of that type the pair has. If the type is weighted by value, its score is scaled by the pair’s total of that value. Contributions combine so that more types pull harder, but never past full strength: two types at 0.5 give 0.75. Every type at 1 gives the same map as All connection types with Count once. One type at 1 gives the same map as One connection type. That’s why switching to Custom mix doesn’t move anything until you change a score.
Small and missing values
Section titled “Small and missing values”A connection that counts always pulls a little, however small its value. A £500 grant beside a £5M one still holds its pair together. A missing or unreadable value counts as a small one, unless you tick the box to leave connections without one out. A recorded 0 is a value.
Line thickness
Section titled “Line thickness”With Show connection strength as thickness on, each line’s width shows its pair’s strength in real units: a count, a total value or a score. The weakest line on the map gets the normal width, and the strongest gets the Thickness range times that. Lines in between are spread evenly, or on a Log scale, where each tenfold step adds the same width. A log scale makes a £5K grant visibly thicker than a £500 one, even next to a £2M grant.
Thickness uses the real values. It doesn’t include the minimum pull that keeps small values connected. Hiding connections with the Visibility filter doesn’t change other lines’ widths.
Bridges
Section titled “Bridges”Communities are found from every connection that counts toward the layout. They’re not affected by the Visibility filter. The calculation is deterministic, so the same map always gives the same communities. Communities are coloured largest first. The ninth and later share one grey-brown colour, as do nodes that aren’t in any group.
Betweenness measures how often a node sits on the shortest route between other nodes. Stronger connections count as shorter routes. Betweenness is worked out from the connections still visible, so filtering changes node sizes. A node with high betweenness is a bridge.
Node size is relative to the map’s strongest bridge, which is drawn largest. A node grows with its share of the highest betweenness score on the map, so differences show even on a map where every score is low. The Analysis section in the Details panel shows the node’s own score.
Each community’s highlighted bridge is its member with the highest betweenness, or its best-connected member if nothing in it bridges anything.
In the Bridges layout, each community is pulled towards its own centre, and connections between communities pull more weakly over a longer distance. That’s what separates the communities into islands.
Capabilities
Section titled “Capabilities”Each cluster has an invisible centre that pulls its nodes towards it. In Overlap, each tag’s hub pulls on the nodes that have the tag, and a node with several tags is pulled towards the middle of their hubs. Each hub in turn moves towards the middle of its nodes, so tags that share nodes end up close together.
Similarity compares the sets of chosen tags that nodes have. Two sets are as similar as the share of their tags they have in common: Digital + Research and Digital share one of two tags, so they’re half alike. It then finds groups of sets that are more alike each other than the rest, the same way Bridges finds communities. Each cluster is named after the tags at least half its nodes have, or its commonest tag if none reaches half. “Mostly” means not every node in it has exactly those tags.
Leaving nodes out
Section titled “Leaving nodes out”Nodes left out by the Nodes section are removed before anything is worked out. The layout, the connection weights (which are relative to the map’s strongest pair), Bridges’ communities and betweenness, the Capabilities clusters, and the size and colour scales all come from the nodes that are left. A connection to a left-out node goes with it. Connections a step in a reveal leaves out are removed the same way, and so are the nodes that leaves with no connection, if the step asks.
The Visibility filter’s node conditions work differently: they change none of these. Greyed-out nodes keep their place and size, and only Bridges’ betweenness changes, as it follows the connections still visible.
Faded nodes
Section titled “Faded nodes”A node is greyed out when the Visibility filter’s node conditions don’t pick it out, or when it has connections but none of them is visible, because the filter or Draw the map from hides them all. A node with no connections at all never fades for that reason: nothing hid anything of its. Selecting a node highlights only the neighbours it has a visible connection to.
A connection between a node the filter picks out and a greyed-out one stays drawn, at half strength.
How the layout settles
Section titled “How the layout settles”The layout starts from the nodes’ positions on the Edit canvas and runs for a fixed number of steps, so the same settings always settle the same way. Unplaced nodes are positioned by the layout itself.
When you change a setting, the layout carries on from where the nodes are, so you can see what the change does. The result can differ slightly from a fresh layout with the same settings. Re-run layout starts again from the Edit positions, which is the layout a saved view shows when it’s loaded, unless you’re stepping from one step of a reveal to the next.
A view can change the numbers each mode’s layout runs on (how long links are, how hard nodes push apart, how strongly they’re drawn to the middle, and Bridges’ and Capabilities’ own pulls), on the Forces tab of the raw view parameters panel (Shift + Alt + P). A change carries the layout on to its new shape, and saving the view keeps it.
Whenever the layout changes, the map glides to it: nodes move into place, colours and sizes blend, and nodes and connections that leave or join fade out or in. In Bridges the glide waits a moment while each node’s bridging score is worked out.
On maps of more than 600 nodes, the soft shadow under each node is turned off to keep drawing fast.
Image export
Section titled “Image export”PNG export draws the canvas at a higher resolution and marks the file as 300 DPI, the density print work expects. If the browser can’t make an image that large, the export is still marked 300 DPI but is physically smaller, and you’re told.