Social networks and/or professional networks?
by Boris Jardine
What kind of network are we looking at in Tools of Knowledge? Which tools are appropriate and what can we find out?
Throughout the Tools of Knowledge project we have used network analysis, especially in the app Gephi, as a way of investigating our data. Representing all of the instrument makers in the SIMON database as nodes, and using professional and social relationships as edges, we can look at the instrument trade in a completely new way:
As abstract as it is, I agree with Ruth and Sebastian Ahnert when they say that images like this are powerful: were are ‘seeing’ a vast historical phenomenon, i.e. the entire trade of scientific instrument making in Britain from the early-modern period up to the 20th century. To use an apt metaphor, are using a new tool to glimpse and otherwise unseen world.
But there’s relatively little we can do with this on its own. We have already used two algorithms to give structure to the network (‘ForceAtlas’, to give the network shape, and ‘Giant Component’ to isolate the core group of connected makers). Now we can add some graphical elements to identify key makers:
Resizing the nodes (makers) on degree (number of connections) singles out one maker from all the others: this maker is at the centre of the network and is very well connected. Anyone familiar with the history of instrument making will not be surprised to find that this is Jesse Ramsden (1735–1800), one of the leading instrument makers in London in the second half of the 18th century.
Any way you slice it – using traditional or computational analysis – Ramsden is important: for the number and status of his clients, the range of instruments he made, for his technical innovations (especially in scale division), and for the overall quality of his work. But his prominence is also a clue to one of the problems with this kind of analysis, because Ramsden happens to be one of the very few instrument makers whose workshop is at all well known. Anita McConnell’s masterful biography has revealed more about Ramsden than we know about any of his contemporaries. Yes he was important – but we want to know who else was important, and Ramsden’s mark on the historical record is simply too big, obscuring others.
To get around this, and to deepen our analysis, we need to think a little more about what we are looking at in the SIMON data. This is not a traditional ‘social network’, made up of people who know each other at any one time. Instead it is a vertical or diachronous professional network, where most of the edges represent a legally binding relationship between master and apprentice.
If these are largely professional relationships, then we can’t really talk about ‘influence’ or ‘importance’ in a network in the same way as we would if all the nodes were connected by, for example, exchanges of letters or personal contacts. When you are bound to one master as an apprentice, you (usually) don’t move to another master – it’s a one-shot deal. And having lots of apprentices at any one time may not be important as having one apprentice to whom you pass on all of your trade secrets. Indeed, in some guilds it was only legal to have one apprentice at a time, severely limiting inferences we can make about workshop size based on the number of apprentices.
So we need to work with this diachronic and dynamic feature of our data. One way to do this is to consider the master-apprentice system as a matter of information flow rather than social influence. Masters taught their apprentices how to make instruments, and in fact for hundreds of years this was pretty much the only way to learn how to make instruments at all (whatever instrument manuals might suggest…).
What does our network look like if we isolate only master-apprentice relationships?
In addition to isolating just this one kind of relationship, here are looking at nodes that are resized to show ‘Eigencentrality’ (proximity to well-connected nodes) and shaded from cold to warm on ‘betweenness centrality’ (how many shortest-paths run through each node).
What we are looking for here are makers who score highly on betweenness, and relatively low on Eigencentrality (smaller red circles):
Note that Ramsden has gone altogether. His importance was largely owing to the large number of employees he’s known to have had. Now we see a group of relatively little known makers stand out from the crowd:
- John Neale, scale-maker, active 1697–1739
- Samuel Saunders, mathematical instrument maker, active 1708–1743
- Daniel Thompson, scale-maker, active 1725–1745
- John Swithin, scale-maker, active 1739–1764
- Robert Vincent, scale-maker, active 1751–1793
- Samuel Bassingham, scale-maker, active 1791–1828
- Nathaniel Stanley, scale-maker, active 1816–1822
This is very far from being a well-known group of instrument makers, yet we can say with confidence that they played an important role in passing on craft know-how from generation to generation. All but one of these craftsmen specialized in scale-making – so they do not fit our ideal type of instrument maker, who by the 18th century would be someone offering ‘mathematical’, ‘optical’ and ‘philosophical’ instruments.
Now we would like to know how this group of quite specialized makers relates to the larger cohort of instrument makers – though we will have to eliminate the confounding Jesse Ramsden by allowing only makers who began work up to 1762, Ramsden’s earliest date:
Again we are seeing size representing Eigenvector centrality and colour representing betweenness. Two things jump out in this network: the central group around Richard Cushee (active 1729–1732), and a new group of very well placed makers like Christopher Stedman (active 1747–1774), George Graham (active 1713–1751), and Henry Gregory (active 1744–1772). There is very little overlap with our group of scale-makers at all, and this gives us some important information: the scale-makers were important only when we isolated master-apprentice relations, strongly suggesting that those relationships remained significant to scale-makers long after they had ceased to play such a central role in the community of instrument makers considered more broadly.
Cushee (globe-maker), Graham (clockmaker), Stedman and Gregory (both mathematical instrument makers) are much better known than any of the scale-makers. Ahnert and Ahnert point out that we want at least some of our analyses to confirm existing suppositions and support prior research, as this helps build confidence in network analysis itself. Yet in that case we must note some surprising absences: no Elias Allen, Henry Sutton, Thomas Tuttell, George Adams, Benjamin Martin…
With one positive finding (the importance of master-apprentice relationships amongst scale-makers) we can admit defeat on other fronts. It is hard to interpret the importance of Cushee et al. until we know more about the exact kinds of connections we are looking at. One obvious conclusion is that we need a range of techniques in order to back up our findings. Within network analysis we need to continue to facet on a range of different edge-types (as we did to identify the scale-makers), and probably also limit by time period (as we did to remove Ramsden).
But we also need to combine this technical analysis with traditional methods: are there histories of scale-making that we can use to test our findings? Will further reading about Cushee, Graham et al. explain their special place, above other more famous makers? Network analysis combines power and precision with inherent uncertainty, but both of these are ultimately strengths as they push us to further study.
Leave a Reply