Networks

Last revised 14 October 2024.

Mary Hamilton's life (1756 – 1816)

We map social networks in five unequal periods of Mary Hamilton's life, in order to visualise who she principally interacts with, and their interrelationships:

  • Adolescence (1758 – March 1777, including a few items addressed to her parents during her childhood)
  • Court (April 1777 – November 1782, when she was employed by Queen Charlotte as a sub-governess to the princesses)
  • Clarges (December 1782 – July 1785, when she shared a house in Clarges Street, had a hectic social life, and kept diaries)
  • Marriage-N (August 1785 – January 1797, when her marital home was in the north of England)
  • Marriage-S (February 1797 – July 1837, when her marital home was in the south, including a few items postdating her death in 1816)

Results

The visualisations display multiple interconnections among a few dozen of the most significant persons in that period. Unsurprisingly, given her ubiquity in our source materials, Mary Hamilton plays a disproportionate part (John Dickenson too in the last two periods), not least because she often provides the shortest path between two other nodes, a bridging role that was rated highly in the processing.1

For this exercise we did not use the more than two thousand familial relationships recorded in our personography database during the project; the data fed into the Gephi mapping software took no account of family ties. What is striking is that many husband-wife, parent-child and sibling pairings nevertheless become apparent in network visualisation, where each node colour represents one densely interconnected cluster:2 thus in the Clarges period we see the Dawsons, Glovers, Jacksons, (Frederick) Hamiltons, Veseys and Wakes. Many of the Bluestockings are grouped in another cluster. Much the same is true in other periods of Hamilton's life, supporting the claim that plausible networks can be derived from document metadata and transcriptions alone, as explained below.

 

Methodology: A sketch

The information for network visualisation is wholly derived from documents in the archive, using an innovative, reproducible methodology that requires little or no close reading, subjective judgement or consideration of external databases or collections. We combine metadata from all correspondence in the archive, whether or not transcribed, with information derived from people mentioned (‘mentionees’) in transcribed correspondence and diaries. The strength of links between individuals is derived from the frequencies of author-recipient, author-mentionee, recipient-mentionee and mentionee-mentionee co-occurrences, counted once each per letter or day's entry. The contribution of each type of frequency is weighted by careful sampling. By going beyond the metadata and utilising textual mark-up already present in the transcription files, we can begin to show relationships not involving Mary Hamilton or her husband, who between them are letter-writer or addressee in the vast majority of correspondence. The technique will be described in full in

Denison, David & Tino Oudesluijs (in progress), 'Reconstructing Mary Hamilton’s social networks'. To appear in Sophie Coulombeau, David Denison & Nuria Yáñez-Bouza (eds.), Mary Hamilton and her networks: Gender, sociability, manuscript, c.1740 – c.1850.

 

There are still questions to be decided. We have experimented with omitting edges under a certain weight in order to minimise the risk of spurious ties between nodes. More recently we have also used the measure ‘closeness centrality’ to restrict display to the most salient nodes. What are the appropriate thresholds? We aim to strike an appropriate balance between comprehensiveness on the one hand and visual clarity on the other.3 Which nodes to remove on principle, such as persons no longer living? What then of the occasional living person known purely by reputation (e.g. Napoleon, much discussed by network members in the marriage-S period)? It is hard to distinguish such a figure systematically from a public figure like George III, likewise very much ‘in the news’ but also known personally to a number of members.


 

Notes

1 The measure is called ‘betweenness centrality’.
2 Such a cluster is called a ‘modularity class’.
3 The Gephi files provided for download contain separate datasets for network maps with different numbers of nodes.

Follow us on Twitter!