Network analysis in corpus: from a tie table to a graph and its measures

In interviews, people constantly mention other people. “A guy from my home town got me onto a building site.” “My brother helped me find a place to live.” “My flatmate found me my first job.” These mentions can be turned into a network if they are recorded in a consistent form. Here is how to do it in corpus, and what to keep in mind when interpreting the result.

Step 1. A three-column table

The starting point is a three-column table. One row, one tie:

individual	tie type	individual 2
P-014	same home town	P-022
P-014	helped find work	P-031
P-022	relative	N-02
P-031	same home town	P-027

A few rules that will save you an hour at the next step:

• People are behind codes. Participants get their own codes (P-014), and people who were only mentioned get separate ones (N-02, “P-022’s brother”). There should be no names in the tie table: network pictures have a habit of ending up in presentations.
• The same thing, written the same way. “P-14”, “P-014” and “P 014” are three different people to any software. The same goes for tie types: agree on a short list in advance.
• Direction, on purpose. “Helped find work” is a directed tie: the first column is the person who helped. “Same home town” is undirected. If in doubt, record the direction: it is easier to drop it later than to reconstruct it.

Step 2. Upload

In corpus, you can upload the table as an XLSX or CSV file, or simply paste it from the clipboard. Headers are recognised in Russian, English and French, and a template with the right columns can be downloaded straight from the network page. If you have an adjacency matrix rather than a list of ties, that works too.

If the codes in the table match the codes of the study’s participants, the nodes automatically pick up their attributes: age, region, interview wave. That comes in handy when you want to colour the network by group.

Step 3. Lay it out

A layout is a way of placing the points on a plane. There are several, and they answer different questions:

• force-directed: connected nodes attract, unconnected ones repel; dense groups stand out clearly;
• Kamada–Kawai: tries to make distance in the picture resemble distance in the network;
• circle: all the nodes around a circle; for small networks where you need to see every tie;
• by group: nodes from the same group are placed together, for example by region of origin.

You can drag nodes around by hand, and their positions are saved. That is useful for an illustration in a chapter, but this is the moment to stop and say something important.

a picture is not a finding, it’s a reason to look at the table.

A layout can draw a “cluster” where there is none, and hide a tie behind another node. Whatever you see in the picture, check it with numbers.

Step 4. Run the numbers

A few measures worth knowing by sight:

• Density: the share of ties present out of all possible ties. In a network of 20 people, 190 undirected ties are possible; if there are 19, the density is 0.1.
• Degree: how many ties a node has. These are the most obvious “centres” of the network.
• Betweenness: how often a node lies on the shortest path between others. In migration networks, this is often the person who arrived first and “knows everyone”. Work, housing and news pass through them, and without them the network would fall apart.
• Communities: groups that are more closely connected internally than to the rest.

And the main caveat. A network built from twenty interviews is not the network of a community; it is a network of what was told. It contains the people who were remembered and whom the interviewees chose to name. A missing node here is not a random gap but a sociological fact, and one worth noting in a memo.

Step 5. Save it so you can find it in a year

You can export the picture as PNG or SVG (SVG for print and editing), copy it to the clipboard for slides, or save it as a snapshot in the gallery or the study’s field archive. A snapshot remembers its date, so a year from now it will be clear which version of the network is in chapter 3.

For further analysis, the network can be exported to GraphML and GEXF, to a matrix and to an R script. For example:

library(igraph)
g <- read_graph("network.graphml", format = "graphml")
sort(betweenness(g), decreasing = TRUE)[1:5]

Checklist

1. Three columns: individual, tie type, individual 2.
2. Codes instead of names, consistent spelling, a short list of types.
3. Upload, and check that the nodes match the participants.
4. Lay it out two ways and compare.
5. Calculate density, degree and betweenness.
6. Write a memo: what is missing from this network and why.
7. Save a snapshot to the archive.

Every step — table upload, layout, snapshot — is kept in the study, so a year later you can see which data the network in your chapter was built from.