Margaret Arfaa, with Cultures of Knowledge, using metadata supplied by Louisiane Ferlier (The Royal Society, London)
The following is a user guide to the visualization tool the ‘Royal Society Genealogy Explorer’, a tool designed for researchers to help them study the social dynamics that influenced The Royal Society’s formation and early years.
How to Use:
A set of social relations are represented as a graph with each node representing a Fellow of the Royal Society. The construction of the graph is simple: if a member proposed another member for election they are connected via a directed edge (→) that points from the proposer to the proposed. There are two methods of exploring the graph. The first is to click on a node, and the second is to select a Fellow using the drop-down menu. Once you have clicked on a node or selected a Fellow, all the ‘descendants’ of that Fellow will be highlighted in yellow, and all his ‘ancestors’ will be highlighted in red. A first-generation descendant is anyone that the selected Fellow proposed, and a first-generation ancestor is anyone who proposed that Fellow for membership. A second-generation descendant is anyone proposed by the selected Fellow’s first-generation descendants, and so forth for subsequent generations of descendants. A second-generation ancestor is anyone who proposed the selected Fellow’s proposers, and so forth for subsequent generations of ancestors. Please note that not all Fellows represented in the graph will have ancestors and descendants: because the graph only covers proposer data to 1740, there are some Fellows for whom the proposer has not been recorded, and some Fellows who were founding members and therefore required no proposer.
Once a selection has been made via the graph or the drop-down menu, the information panel will display statistics, linked data, and record information. These fields are as follows:
- Royal Society Authorized Name: The officially listed name format in the Royal Society database of Past Fellows.
- Election Date: The date that Fellow was elected to membership of the Royal Society.
- Royal Society ID: The internal code that the Royal Society uses to identify the Fellow in its Past Fellows database.
- Profile (Royal Society): A link to the Fellow’s profile in the Royal Society’s Past Fellows database.
- EMLO Person ID: The unique record identifier that Early Modern Letters Online [EMLO] uses to identify this Fellow within its union catalogue database.
- Profile (EMLO): A link to the Fellow’s EMLO record.
- Number of Proposers: The number of pre-existing members to propose the selected Fellow for membership.
- Number Proposed: The number of members for whom the selected Fellow is recorded as a proposer.
- Genealogy Depth: The longest chain of descendants that emanates from the selected Fellow.
- Total Descendants: The total number of descendants of every generation that emanate from the selected Fellow.
- Total Ancestors: The total number of ancestors from whom the selected Fellow is descended.
- Internal ID: The ID that the visualization app uses to identify the Fellow within its own code and data repository.
- Generation k Descendants: A list of all the kth-generation descendants of the selected Fellow.
- Generation k Ancestors: A list of all the kth-generation ancestors of the selected Fellow.
The drop-down menu has some additional features. You can select a presidential or secretarial term to see all Fellows elected during that term or you can simply select ‘Founders’ to highlight all the founding members included in the graph. You can also select ‘Domestic Fellows’ for Fellows listed as being English or British (none was listed as Scottish or Welsh), ‘Foreign Fellows’ for Fellows with nationalities other than English or British, or ‘Fellows with no Recorded Nationality’. If you select a group category, the information panel on the left side will list the group name, the number of Fellows in that group, and the names of all the Fellows in that group. A term of office is calculated to exclude the day of election of the officer and include the day of transfer to the incoming officer.
It should be noted that the years accompanying each Fellow’s name are not necessarily the years of their birth and death. They are the years that the Royal Society puts under the heading ‘Dates’ in the Past Fellows database. In some cases, the dates represent the lifespan of the Fellow, for some they represent only their time within the Society, and some only have a single date which is usually either the election or death year. Please see the Society database pages on particular Fellows for detailed chronological and biographical information.
All the underlying data are available for download at the bottom of this page.
Contextualizing the Data:
Although visualization can create the illusion of comprehensive, definitive data, the dataset should be considered in context. It is worth noting that the period for which we have Fellows with known proposers in the dataset ends in the year 1730,* a decade before the end of the period covered in EMLO by the Royal Society’s Early Letters catalogue. There are also some years where a higher proportion of proposer entries are missing than others, as well as years for which there are no proposer records at all, even before 1730.
* There was a single exception recorded within the dataset in the year 1758 for Francis Hastings (1729–1789). However, although potentially worthy of exploration, this entry falls outside of EMLO’s scope so was included in the graph visualization but excluded from the analysis.
Fig. 1: ‘Number of Fellows Elected Per Year’ displays all elections for each year counted in blue and elections with a record of the proposer counted in red. An average of 14.68 Fellows were elected each year, a mean of 5.26 of which would have been recorded, with standard deviations of 13.65 and 5.42, respectively.
Even excluding founder members, honorary members, patrons, and royal members, none of whom, due to the nature of their membership, would have been a part of the normal selection process, only about 36% of Fellows elected between 1660 and 1740 have recorded proposers. Given the large proportion of missing data, any analysis of the early Royal Society derived from these data should be considered with caution. However, the fluctuations in record-keeping over time in and of themselves may provide interesting sites for analysis. Did some secretaries have better archival practices than others? Did the Society change its record-keeping practices post-1730? And, if so, why?
Fig. 2: ‘Proportion of Fellows with Known Proposers’
displays in red the percentage of elections
for which a proposer was recorded.
Additionally, decisions were made in cases where there is proposer data, but the identity of the proposer is unclear or multiple possible identities are listed. Please see the data-cleaning notes (Appendix C) for a record of all such decisions that were made and why. All data are from a printout created by Louisiane Ferlier received from the Society by EMLO on 1 May 2020 and transformed from an XML file to an Excel file by John Pybus. For the most up-to-date information on each Fellow, click the links in the information panel to access their current Royal Society database profile or their EMLO person record. Dates indicating secretarial and presidential tenures were supplemented from The Record of the Royal Society of London for the Promotion of Natural Knowledge (4th edn, 1940).
Fluctuations in elections and the way records were kept can also be understood in terms of the leadership of the Royal Society. The following graph overlays presidential tenures to help visualize how fluctuations in the growth of the Royal Society and in record-keeping may be observed under different leadership.
Fig. 3: The ‘Number of Fellows Elected Per Year’ are displayed with the terms of Royal Society presidents superimposed.
Looking at the dynamics of recruitment across time in relation to the presidents of the Royal Society, rather than only the total successful recruitment during each tenure, reveals the different patterns of recruitment and their results. For instance, although William Brouncker (the Society’s first president from 1663 to 1677) had the most dramatic spike in elections, Isaac Newton (president from 1702 until his death in 1727) outpaces him in total elections due to his long-term pattern of regular elections. In this context, Brouncker’s success could be interpreted as a flash in the pan as people flocked to the newly established Royal Society, differing from the steady interest that Newton was able to sustain.
Fig. 4: ‘Number of Fellows Elected Per Presidential Term’ displays both the total number of elections in each president’s term (blue) and the number with a recorded proposer (red) in descending order. On average, presidents oversaw 88.62 elections during their tenure, a mean of 32.08 was recorded, with standard deviations of 121.72 and 49.90, respectively.
Above is a more compact visualization of each president’s recruitment and record-keeping, ordered by number of successful elections rather than chronology. The three presidents with the most elections, Newton, Brouncker, and Hans Sloane (Newton’s successor and president from 1727 to 1740), also have the three longest tenures of any president. This should be considered in cases where elections are used as a measure of a president’s effectiveness, but the raw number of elections could still be a strong indicator of lasting influence over the Society’s continued shape and formation.
Fig. 5: ‘Number of Fellows Elected Per Secretarial Term’ displays the total number of elections (blue) and the elections with a proposer record (red) that occurred during each secretary’s tenure arranged in descending order. Note that non-consecutive periods in office were counted as separate terms and all term lengths were artificially capped in 1740. On average, secretaries oversaw 95.08 elections during their tenure, a mean of 35.76 of which were recorded, with standard deviations of 104.39 and 40.21 respectively.
Recruitment and record-keeping can and should also be considered relative to the Society’s secretaries. It should be noted that John Machin, Cromwell Mortimer, and Hans Sloane had their terms artificially capped at 1740 in order to restrict the scope of the analysis. No Fellows who were elected during their terms after the year 1740 were counted. The Society also always had two secretaries at the same time, so the graph of elected Fellows per secretarial term essentially double-counts member elections. This fact means that an underactive secretary might be falsely credited with influence using just raw elections if he shared some or all of his tenure with an overactive counterpart.
Terms exclude the first day of office and count the day of transfer of power to the next officer. Not doing so would lead to some minor absurdities, for example Henry Oldenburg being elected to membership within his own tenure as secretary.
Fig. 6: ‘Number of Successful Elections Proposed Per Fellow’ displays the top fifty Fellows in terms of number of elections in descending order.
It is evident that some members made many more successful proposals than others, with a relatively small number of Fellows making many more than the vast majority. Considering the whole population of Fellows who made any recorded proposal, the typical Fellow would have made 3.20 recorded proposals with a standard deviation over the whole population being 5.60. However, using this as a proxy for active membership is confounded by at least three factors. The first is the missing records and the uneven distribution of missing records across years, in which different members might have been more or less active (not to mention the complete lack of records after 1730). The second is fluctuations in broader interest in the Society that might have affected recruitment efforts of all members in a given period regardless of their individual activity level. The third is that there is no way of knowing whether some Fellows might have been systematically left unrecorded as proposers for one reason or another.
The following visualizations attempt to compensate for the first two confounders.
Fig. 7: ‘Proposal Z-Score Comparison by Election Year’ displays each Fellow’s z-score for true proposal records (red) and for an artificially boosted version of their proposer records (blue) by their election date converted to a fractional year.
The figure above visualizes the z-scores (number of standard deviations above the mean) of each recorded proposer. In an attempt to mitigate the effect of the overall climate of interest or disinterest in the Society, the z-scores were calculated locally, using a sliding window of up to ten proposers with the closest election date to the selected proposer both in the future and in the past (resulting in mean pools of up to 21 and no lower than 11, counting the selected proposer).
This window is temporally flexible by design. In lower periods, it is reasonable to suspect that Fellows would have been more likely to socialize across ‘generations’ and the notion of a peer window should encompass a temporally broader selection of Fellows, in terms of election year. In high recruitment years, it is possible that socialization across generations would have been less common, with a larger social pool facilitating greater temporal stratification. Those elected at the same or a similar time might be expected to socialize together, with new entrants sticking together and potentially being closer to each other in age than more senior members.
The ‘boosted z-score’ is an attempt to compensate for the first confounder, the missing records. Each boosted proposal count was calculated with the equation
where Y is the collection of years in which a given Fellow is recorded as a proposer, p_y is the number of proposals that Fellow is recorded as having made in a particular year, and π_y is the ratio of election records with a proposer to the total elections in the same year.
Considering only each Fellow’s recorded proposals, the following Fellows are considered outliers (z-score > 1), substantial outliers (z-score > 2), and extreme outliers (z-score > 3), respectively.
Extreme Outliers: John Wilkins, Seth Ward, Isaac Newton, Hans Sloane, William Jones, William Rutty, John Gaspar Scheuchzer.
Substantial Outliers: Robert Hooke, Robert Southwell, William Derham, Martin Folkes.
Outliers: Henry Oldenburg, John Hoskyns, John Chamberlayne, Richard Mead, Andrew Tooke, William Stukeley.
Although his work only extends through the year 1700, Michael Hunter in his 1982 monograph, The Royal Society and its Fellows 1600–1700: The Morphology of an Early Scientific Institution (for publication details, see the ‘Sources’ section below), identifies his own list of members comprising what he calls ‘the Society’s active nucleus’, of which he documents the evolution over time. Some of the members that Hunter identifies as active at least for some time are reidentified as outliers above: Wilkins, Ward, Hoskyns (‘Hoskins’ in Hunter), Southwell, Hooke, Oldenburg, and Sloane (pp. 119–21). However, many were not identified.
If the boosted scores are used, Southwell and Folkes are promoted to the extreme outlier class, Ward and Scheuchzer are demoted to the substantial outlier class, Hooke is demoted to the outlier class, and William Brouncker, Robert Moray, John Hadley, Edmund Halley, and John Woodward are added to the outlier class. Hadley, Tooke, Oldenburg, Mead, and Chamberlayne are not considered any sort of outlier at all. Some of the Fellows who gained outlier status when the boosted z-scores were used we also identified as part of the ‘active nucleus’ by Hunter: Brouncker, Moray, and Woodward (pp. 119–21).
Due to the restricted time-window that Hunter examines, it is not possible to make direct comparisons between his ‘active nucleus’ and the proposal outliers. Nevertheless, it is notable that all the Fellows whom Hunter identified as being particularly active were not turned up by either set of z-scores. This result could be a limitation of the measure, since it accounts only for proposals and ignores other activity, or it may be capturing activity at a higher threshold than Hunter sets (see p. 30 for Hunter’s definition of the ‘active nucleus’).
This analysis is intended to demonstrate one of many ways missing data may be silently affecting our ability to use these results as a proxy for activity and influence on the future of the Society. And, as stated previously, the boosting method assumed that all proposal records would have been equally likely to be recorded. The possibility of uneven or consistently biased recording practices may be skewing the results in unknown ways.
Fig. 8: ‘Influence (R) Metric Z-Score Comparison by Election Year’ displays each Fellow’s true influence score (red) and boosted influence score (blue) arranged by election date converted to a fractional year.
The second visualization describes the z-scores of each proposer on a slightly more complicated metric. The metric is a measure of persistence of influence. Each measure was calculated using R = d ⁄ p where p is the total number of proposals made by a fellow and d is the total number of descendants they have in the proposal network. R represents the persistence of a member’s indirect influence on the composition of membership relative to his direct action. Essentially, it is meant to help distinguish members who simply elected many other members from those who effectively shaped the long-term social fabric and composition of the Society, which would have likely shaped its scientific practice and interests.
The figure above displays z-scores of each proposer’s R which were generated using the same sliding-window technique to generate localized mean pools as described previously.
Boosted R metrics were calculated for each proposer by using their boosted proposal score as p and adding each descendant’s boosted proposal scores to get d. From visual inspection alone, it is evident that boosting significantly affected the z-scores of some Fellows. This method is a relatively crude way of accounting for the missing data, yet it demonstrates the weaknesses that the missing records will inevitably introduce into any analysis relying on these data.
Extreme Outliers: Thomas Pellet, William Stanley, Henry Oldenburg, John Keill, Martin Lister, Claude Amyand.
Substantial Outliers: Jonas Moore, Theodore Haak, Paul Buissierm, William Stukeley, John Diodate, Francis Clifton, Robert Hooke.
Outliers: Seth Ward, Philips Glover, Samunda Isaac de Sequeira, Thomas Povey, Henry Heathcote, William Croone, William Brouckner, Peter Desmaizeaux, Samuel Tuke.
With boosting, Seth Ward, John Gaspar Scheuchzer, William Jones, Robert Hooke, Hans Sloane, and William Stukeley were promoted to the extreme outlier category. Thomas Pellet and John Keill were demoted to the outlier category, and William Stanley, John and Claude Amyand were no longer considered outliers. The substantial outlier class was left completely empty. Jonas Moore and Theodore Haak were demoted to the outlier category, while Thomas Povey, William Stukeley, John Diodate, Francis Clifton, Isaac de Sequeira Samuda, Peter Desmaizeaux, Samuel Tuke, William Croone, and Henry Heathcote were no longer considered outliers at all.
Again, there was some overlap with Hunter’s ‘active nucleus’, including William Brouncker, William Croone, Seth Ward, Henry Oldenburg, Thomas Povey, Theodore Haak, Robert Hooke, Hans Sloane, and Martin Lister (pp. 119–21).
While the overlap with Hunter’s ‘active nucleus’ is encouraging, indicating that the R metric is able to detect at least some type of influence or activity, the drastic changes that applying boosting induced in the categories indicates that the missing records are likely a deleterious confounder. Unless there is a way to compensate for the missing data, more sophisticated methods of network analysis (e.g., community detection and eigencentrality measures) are likely to be similarly hampered.
Despite the limitations, the ‘Royal Society Genealogy Explorer’ still provides a valuable perspective for exploring relationships and group dynamics during the early years of the Society and serves as a model for visualization and analysis of recruitment networks. Looking not only at how many times each Fellow is recorded as a proposer but also at the downstream effects through the network of descendants allows for examination of a Fellow’s indirect, in addition to direct, impact. Furthermore, for Fellows who remained outliers both before and after boosting on the proposal or influence metrics, there is significant reason to have confidence in those results.
Hunter provides a more holistic treatment of the Society and its members, considering many more datapoints than simply proposals. However, as the title indicates, The Royal Society and its Fellows 1600–1700 covers data only to 1700 and does not provide a way of visually exploring the Society’s ‘family trees’. This Explorer, thanks to improvements in visualization and webhosting technology, is able to provide an interactive exploration experience and allows for multi-generation tracing of potential influence and connections.
The features that highlight Fellows with a foreign or domestic designation allows one to visually explore the distribution of foreign members across different proposers and genealogies. It also makes it easier to explore the composition of a particular Fellow’s ancestors or descendants with an intuitive visual interface that provides linked archive data both to the Society’s own database and to EMLO, which facilitates easy oscillation between network exploration and archival deep dives. The missing records make the proposal network unsuited for systematic analysis aimed at making general claims about the early Society but, hopefully, the Explorer will aid historians in investigating the direct and indirect relationships that shaped the early Royal Society by allowing them to view its social dynamics from a new perspective.
Sources:
The Record of the Royal Society of London for the Promotion of Natural Knowledge (4th edn, 1940).
Hunter, Michael, The Royal Society and its Fellows 1660–1700: The Morphology of an Early Scientific Institution (Buckinghamshire: British Society for the History of Science, 1982).
The Royal Society. Search Past Fellows.
Acknowledgements:
The greatest thanks for this project go to Miranda Lewis, EMLO’s editor. She supported every aspect of the project and guided me through working to build a tool for the exploration of material that was unfamiliar to me and far from my specialism.
The data procurement and preparation for this project are credited to Louisiane Ferlier (Digital Resources Manager at the Royal Society) and John Pybus (Developer, EMLO, at the Oxford e-Research Centre, Oxford). Louisiane Ferlier provided the data from the Royal Society’s own files in XML format and John Pybus migrated the data from an XML to Excel format.
John Pybus also ensured the Royal Society Genealogy Explorer’s successful integration into the EMLO text page.
Dr Philip Beeley (Research Fellow and Tutor at the University of Oxford History Faculty) provided expert advice and made recommendations on how to make the tool most useful to historians of early modern science.
Millie Gall (Editorial and Administrative Assistant, EMLO) proofread and suggested edits on drafts of this page.
ChatGPT (versions 4o, o3, o4-mini-high, and 4.1o1) assisted in writing and producing code for the Genealogy Explorer itself as well as the visualizations on this page.
Appendix A: Data Downloads
Copies of the underlying CSV files that support the Royal Society Genealogy Explorer as well as a printout of the original data are available below.
Public_Complete_RS_Graph_Metrics_MetaData.csv
Public_RS_data_cleanMembershipColumn.csv
Public_RS_data_ConfirmedMatches.csv
Public_RS_Graph_Metrics_MetaData_with_Office_Data.csv
Public_RS_Proposer_Data_Cleaned&Graphable.csv
Appendix B: Python Documentation
The Royal Society Genealogy Explorer was coded in Python, version 3.10.9, using the following packages:
- dash (v3.0.4) [software]. Date unknown. https://dash.plotly.com/
- dash_cytospace (v1.0.2) [software]. Date unknown. https://dash.plotly.com/cytoscape
- networkx (v3.4.2) [software]. 2024. https://networkx.org/documentation/networkx-3.4.2/reference/index.html
- pandas (v1.5.3) [software]. 2023. https://pandas.pydata.org/
The python file that runs the Explorer is available for download and inspection here:
Appendix C: Notes on Data Cleaning
The primary challenges for data cleaning were that the XML–Excel translation was such that all of a Fellow’s Society activity was listed in a single cell with inconsistent formatting and that proposer listings needed to be automatically name-matched to person records. The first challenge was to separate each piece of information into its own column, exploiting the relative consistencies of the formatting and manually checking and correcting where inconsistencies caused errors or erroneous entries. The proposer entries were only listed as they were originally recorded and were not linked to authoritative IDs, so simple ID matches were impossible. The names were considered auto-matched if two parts of the constructed name-based ID of a Fellow contained at least two name-parts of the proposer entry and the proposed Fellow was elected between his election date and his death. Any cases of multiple matches, lack of matches, incomplete matches, or inconclusive records were manually inspected and adjudicated.
Detailed notes on manual changes made to the data can be downloaded here:
Notes on Manual Data Changes & Decisions.pdf
Appendix D: Z-Score Tables
Download CSV files containing each Fellow’s proposal, boosted proposal, influence, and boosted influence z-scores here: