The mappings below represent the Twitter traffic generated at two related events this October. These were the Impact Convergence and American Evaluation Society annual conference, one happening immediately after the other and involving many of the same people. Impact Convergence was for people involved or interested in the idea of measuring social value and supporting or financing initiatives on the basis of having measurable social value (my interpretation and wording). The evaluation conference, of course, involved people involved in evaluating the results of policies, programs and investments in mainly social, economic and health/education initiatives. Clearly there is an overlap in interest, and we wanted to see what that might look like. Capturing the Twitter traffic generated by these events was one way to do it.
Here’s a clip of outgoing AEA President John Gargani’s closing address at the AEA Conference. At the 1:30 mark to about 3:25 he explains our Twitter mapping to the plenary audience.
The two interactive mappings below were done first with NodeXL, to obtain the data and then process it to produce a set of connections and metrics. This network data was then read into a D3.js file set up to produce an interactive force-directed visual. “Force-directed” means that the points in the mapping are placed relative to each other depending on how strongly and directly or indirectly they are connected – kind of a “gravity” concept.
The first interactive mapping captures all Twitter messages using the official hashtags and handles of both events. The points move around at first as the force-direction calculations take place and converge to a stable state. This is a large network, so not all of it fits in the pane. I have left it this way rather than “squashing” it all in as what is of interest at the moment is the central group connecting both events (I am in that central group, but only because I was regularly sending out updates on the network). A somewhat larger view is available in the “Portfolio” section of this website.
You will see that some nodes (individuals who tweeted or are mentioned in tweets) are larger than others. This is based on “betweenness centrality”, that is, the extent to which one is a “connector” or “bridge” between others. The colours of the nodes denote membership in “clusters”. Members of clusters are relatively more connected as a group to each other than to the rest of the network. Clusters are usually underlain by commonalities among their members, such as interests or affiliations.
There is a strong central group of players who were attached to both events and helped to connect many others. A good way to see this more clearly is to use the interactive “drag and fix” function of this D3.js visualization. I suggest the following: double-click on the large “aeaweb” node until you see a black outline appear. The node is now “fixed” and can be dragged with your mouse to a position of say, an inch from the top of the map. Then, once things stop moving so much, do the same with the “impact_converge” node, dragging it to a similar distance from the bottom margin. As things begin to settle, you will see, in the middle are between the two fixed points, a set of influential “connectors” spanning both worlds.
The second interactive mapping represents the same network. In this case however, the interactive feature allows you to focus on the immediate connections of any given node by double clicking on that node.
Along with the interactive mappings and their usefulness for exploration of the network, we did an analysis of the extent to which the conferences overlapped in participants’ Twitter conversations.
The Impact Convergence was for people involved or interested in the idea of measuring social value and supporting or financing initiatives on the basis of having measurable social value). The evaluation conference involved people involved in evaluating the results of policies, programs and investments in mainly social, economic and health/education initiatives. Clearly there would be an overlap in interests, and AEA wanted to promote interchange between these communities. AEA naturally also hoped to be able to gauge their success in fostering this “convergence”. Capturing and mapping the Twitter traffic generated by these events as a social network was one way to do it.
If there was no overlap between the Impact Convergence and Evaluation 2016 Twitter traffic, then we would have seen something like the first image: two entirely separate communities.
Instead of two non-overlapping groups, what we did see was the second: intersecting communities, several of which are clearly either primarily about Impact Converge or Evaluation 2106, but also clearly overlapping, and particularly where a number of key network members are concerned. We will discuss what makes them “key” in the next post. The key “both-world” players are grouped between, rather than immediately surrounding, the centres of @Impact_Converge and @aeaweb (the official Twitter addresses for these events).
Understanding this observation and its implications is assisted by thinking of this as a matter of gravity. In this “force-directed” network layout, all nodes are repelled to a degree, but the repulsion is countered by gravitation to other nodes that is stronger the more often the nodes are in contact (via Twitter). Proximity in the mapping, therefore, is a function of connection. Nodes that share many mutual connections will tend to be proximal in the mapping.
Reasonable proxies for the centre of the IMPCON and Eval 2016 networks are the locations of the @Impact_Converge and @aeaweb nodes. Not all tweets included these handles, but many did, and those not referencing them and using the conference hostages of #IMPCON16 and #Eval16 would still tend to be connected to others who did, for one event or the other. Gravity would thus bring them together into the neighbourhoods surrounding these nodes, as is clear in the mapping.
The first degree (immediate) connections to these nodes extend into these neighbourhoods, overlapping in the space between them. The significance of this is further explained next.
Selecting (approximately) the members of the group connected directly to both of the main event nodes, and selecting their first degree neighbours, gives us a “sub-graph” of participants who are very close to either or both events in terms of the sources and destinations of their own communications or those with whom they are directly connected.
You could use this group to represent the main overlap of the events.
Another way of looking at this is by differentiating between clusters,“pulling them apart” as in the final. layout. This makes the clusters, identified in the first mapping by the same colours as in this mapping, more apparent.
The gravity of mutual connection pulls the people in G1 in particular away from both of the event centres, while some of the central overlapping group are still located within the IMPCON-centred cluster.
The same is true of the other smaller clusters as well: while many of their members may be connected to the central event nodes, they are still more strongly connected to the others with whom they are clustered.
Most of the central overlapping group are found where circled in two large and highly interconnected clusters. Being able to zoom in on subgroups like this is handy strategically as part of a communications strategy.