On the 1st of June 2015 the National Archives released a tranche of documents compiled 100 years beforehand by the Dublin Metropolitan Police. Agents of the DMP followed a number of individuals involved in Sinn Fein, the Irish Volunteers, the Gaelic League and the suffragettes and reported on their movements to Dublin Castle. Superintendent Owen Brien would take the previous evening’s reports, compile them into a single document and pass this along to the Under Secretary for Ireland. The reports date from the 1st of June 1915 (covering the events of the evening beforehand, Monday 31st of May) and continue until the 20th of April 1916. 4 days later on Monday the 24th many of the individuals mentioned in these reports took control of buildings throughout Dublin and proclaimed an independent Irish Republic.
After looking through a few of the reports I got the idea of compiling them into an edge list and building a network from them. The full network contains 233 different nodes connected by 3995 weighted edges. An edge exists between two individuals who are reported by the DMP as being seen together. The weight of the edge represents how many times they were seen together. The graphs below were created using Networkx in python and Gephi and I have a number of more detailed jupyter notebooks which I am going to clean up and put on my github.
The full network is too big and has too many weak connections to provide any meaningful insight so I built some smaller graphs by looking at the strongest relationships in the network. Of the almost 4000 edges 2249 of them have a weight of 1, which means that the two individuals the edge connects were reported as being seen together only once. There are only 55 nodes that have at least one edge with a weight greater than 10. At the other end are some very strong connections: Tom Clarke met with Thomas Byrne 106 times and he met with William O’Leary Curtis 105 times.
By trial and error I found that building a network using edges of weight greater than or equal to 25 (ie, where every person had spoken to at least one other person more than 25 times) led to two clusters forming, bridged in the middle by Seán MacDiarmada. Taking these nodes and using them to form a subgraph of the full graph means many new, weaker, connections appear but the overall structure of the previous graph remains.
The first is a star cluster, a single central node (Tom Clarke) with all the other nodes connected to him. It includes a number of Irish Republican Brotherhood members like Seán T. O’Kelly, Diarmuid Lynch, Seán McGarry and Piaras Beáslaí.
The second cluster is heavily connected and is centred mainly around The O’Rahilly and Bulmer Hobson. This cluster represents the upper echelons of the Irish Volunteers who met regularly. The Volunteers didn’t know that the IRB was planning a Rising in 1916 and it’s Chief of Staff, Eoin MacNeill, appears removed from the centre with no strong connections. Éamon De Valera is also removed from this graph, knowing nothing about the plans until he was bought onto the IRB shortly before Rising. The nodes are sized in Gephi by PageRank with larger nodes being more popular. The nodes and edges are coloured using Gephi’s modularity statistic which searches for communities in the network.
This structure, a Clarke community and an Irish Volunteer community, replicates across many different graphs. Taking the full network I calculated a PageRank score for each node, a measure of its popularity. I took the top 50 most popular nodes and built a new graph which better shows the gap between the central nodes in both the Clarke cluster and the Irish Volunteers cluster and how the other nodes are pushed far outside of this.
Arthur Griffith continues to be pushed further away from the centre, being kept in the dark as to what was being planned. Joseph Plunkett appears on the outskirts of the Irish Volunteers cluster; he spent time traveling to Germany and New York in mid and late 1915 and suffered severe ill health in the later stages of the planning. James Connolly, bought into the IRB only in January, appears isolated in the top left, having appeared in reports mainly through Labour and Union meetings.
Clarke’s position in these graphs could be a result of the way the data was collected. Each report begins with an account of those who entered Clarke’s shop that day with an agent assigned to watch it from open to close. Other than J.J. Walsh for the first few months no other private residence or business was given this much attention. We have to remember that these graphs are built from information available to the British authorities at the time. Maybe they missed a lot of meetings which would have given someone else a central position of importance in the graph. Maybe all the people visiting Clarke were actually there to just buy tobacco.
I set out to examine this network to see what type of information the British had, how effective they had been in following the most important people in the nationalist movement and what type of story they could tell with modern social network analytics tools. The appearance of two segregated communities linked by individuals loyal to Clarke and the isolation of figures like Eoin MacNeill and Arthur Griffith is in agreement with what we know today of how the 1916 Rising came about.
I am currently cleaning some jupyter notebooks which include centrality measures and will add those later. I am interested to see how modern data science techniques can be applied to other aspects of Irish history, especially with the amount of data that is now being released. Last year I applied natural language processing to the Anglo-Irish Treaty debates and got some interesting results. If you know of any similar projects please send me on a link.