The major roadblock- actually visualizing the network in a coherent way- was thankfully solved earlier this week in our meeting! Filtering out the Degree by 8 (doing it by 10 also looks good) as well as making sure the nodes and edges matched correctly created something super nice in Observable! Minor changes to that code included taking out the .scaleExtent attribute as was suggested in the Slack, as well as altering the labels so that they are centered in the middle (modifying it to just d => d.x and adding .attr(“text-anchor”, “middle”) to center it.)
The problem becomes one in which it’s difficult to convey the extent of William Simmons’ network in relation to the whole network because of the degree filtering. By filtering out individuals who have less than 6, 8, or 10 degrees, it is the Secretaries of War and the Quartermaster who emerge as the individuals with the most unique correspondence rather than the previous dominator of the list, William Simmons. If I’m understanding it correctly, it’s because the filter removes those individuals entirely and thus does not allow the connections to be made between them and those who remain, lowering their overall degree. So now it’s a matter of determining whether or not the 5-degree visualization (going lower is too messy) is worth it- filtering out by 5 degrees creates a situation where William Simmons has 155 degrees and James McHenry, Secretary of War, has 154. It’s hard to capture that difference in the network because the nodes are sized exactly the same. (Fork of the Graph w/ zoom, individuals with less than 5 degrees filtered out: https://observablehq.com/d/9ae1a24998503a46)
Still, several clerks and accountants still remain on the list, just with a different degrees depending on what’s filtered out (which makes sense, proportionally). So, I think that depending on the versatility of a tooltip, I can highlight the nodes in the visualization to indicate their role. If I were to pair this with a Tableau visualization with the percentage of correspondence these individuals contributed in relation to the whole, I think you could compare the larger network of these clerks and accountants to the filtered-by-degree network, and conclude that the even accounting for the quasi-war occuring between approx. 1798-1800, they still could not be overtaken in day-to-day unique connections. I could include the year slider by including the author_year of the correspondence but I’m also afraid of breaking it all over again, so I’ll do that experiment in a separate notebook! Otherwise, the graphs will be able to zoom and drag for people to explore the network.
To do:
- Draft the short paragraphs that will introduce the topic and War Department Papers project
- Decide on the degree network visualization that works best here- the bigger one (5 Degrees) or smaller, cleaner one (8-10 degrees). I’d still like to include the 1799 force graph once I fix the nodes just because it displays the largest amount of correspondence and it’s fun to move it around.
- Quick Tableau visuals for the percentages (color coordinated)
- New reconstruction of what I want the project to be formatted like- due to the pictures, graphs, and text I want to include, I think just using WordPress might be simplest? But if I’m having trouble with the look of everything when it’s embedded, I’ll probably look closer into the Github blog option, if only to play around with it.
One reply on “Mod 11 Work”
Github will probably be more than what you need for this project–that usually works best when there’s lots of separate pieces where it makes sense for different parts to have their own separate page (like a museum exhibit with different possible paths through the exhibit). If you’re thinking a fairly straightforward linear narrative, a post on the course site should suit the project fine.
Filtering out the low degree nodes before or after the network is created will change how the final network looks–if you filter out the low degree nodes just before exporting graph.json, your final nodes I believe should still have their betweenness centrality calculated with the low degree nodes taken into account. I would try sizing your nodes by betweeness anyway, since that’s a better measure for the question you’re asking than raw degree (betweeness is a measure of “influence” or connectedness in a network). If eg Simmons is just corresponding with a bunch of low degree randos who don’t correspond with anyone else, he’s got a different role in the network than if he’s corresponding with the major players who are also talking to each other. (Think about Anna Peek in my All One People article–she’s got a very low degree but she’s got a high betweeness because she’s the main lynchpin between two otherwise unconnected networks)