X-Rotary (13): Epilogue
This post will close our series of essays devoted to the network analysis of meeting attendance at the Shanghai Rotary Club. This series has explored the value of a multimodal approach to persons in networks. What conclusions can we draw from this experiment?
Our bimodal analysis reveals that the most active participants in meetings (with high degree centrality) are usually key members of the club (officers, directors), whereas less frequent attendants are either ordinary members or occasional visitors. However, we also identified regular attendants who were not club members proper but locally prominent personalities invited as special guests (U.S. trade commissioner) or Rotarian’s wives serving as entertainers (Hawkings, Kwok). Core events are often – but not always – special events. Regular meetings became a central feature in the life of the club by 1930 (Fitch). During the Sino-Japanese war (Harkson), meeting attendance is more sensitive to the level of information available.
The dual projection method confirms the prominence of core members and events, but also brings additional insights. In co-attendance networks (person-person joined by events), we identified three main personal profiles based on nodes degree and betweenness centrality: leaders (high degree, high betweenness), brokers (medium or low degree, high betweenness) and outsiders (low degree, low betweenness). Leaders are most often key members. Outsiders include either occasional guests or less active members, and the majority of women and non-Westerners. Brokers form a hybrid category, mixed in terms of membership status, gender and nationality. An increasing number of women and non-Westerners played as brokers in the 1930s (especially under Fitch’s presidency). We also observed the hypertrophic centralization around the president during Fitch’s term, which we temporarily attributed to his personal character. Extending the time scope of study would help determine whether such centralization emanates directly from the bye-laws of the club or translates Fitch’s personal application of these rules.
The Rotary club of Shanghai itself contains a multitude of small worlds. Hierarchical clustering reveals that the club can be partitioned into 6 to 11 communities. It was more fragmented under Petit’s term, which translates the founders’ efforts to aggregate new members and supporters during its first year of existence. At that time of incubation, the club functioned along a core-periphery model that complexified in the 1930s, as interconnected peripheral communities developed around the central figure of the president (Fitch), and as (information on) meeting attendance scattered during the war (Harkson).
In the networks of multiple attendance, we shifted from node-persons to edge-persons. We observed that special events and serial lectures (around the same topic) have usually more attendants in common than regular, isolated or narrowly-focused meetings. The number of shared participants decreases over time, which reflects the transition from founding to regular meetings between 1919 and 1930, and the growing scarcity of information in wartime (Harkson). Multiple-role networks reveal that the most frequent and enduring role combinations were: president-chair- speaker, guest-speaker and officer-organizer, whereas such combinations as guest-chair, guest-organizer or neophyte-organizer remain incompatible during the entire period. Alterations in these structural patterns reflect either experiments in assigning and performing roles during the first year (Petit), or disorganization and missing information during the war (Harkson).
I will end by pointing out several limitations in this research and potential directions for further research. Although we attempted to enquire changes in network structure over time, our approach remains static. Can we construct truly longitudinal networks instead of successive snapshots in order to follow Rotarians’ trajectories in the long term? Can we rely on dynamic networks to better understand, if not predict, role taking and role shifting over time? For instance, although Mr. Fong appeared only once in Fitch’s network, he was to succeed Fitch as president of the club for the next term. This suggests that regular meeting attendance, active participation in club affairs or high visibility in press reports were neither the only nor the most determining factors in reaching executive positions in the club. Dynamic networks would certainly shed light on the conditions for, and consequences of, performing specific roles within the club.
A more dynamic approach to persons in networks also implies that we investigate Rotarians’ social life outside the club. In this paper, our analysis focused solely on closed networks. Since Rotary’s motto was to serve society, we need to examine how Rotarians interacted not only within the club, but also with the outside world (local society in Shanghai, national and transnational connections). In this approach, outsiders might be elevated to more central positions. By enlarging the scale of observation and redefining network boundaries, members who seemed inactive or peripheral in Rotary’s small world may prove influential contact points with the outside world. Reconstructing Rotarians’ ego networks may also help better understand their behavior as club members and illuminate their position as leaders, brokers or outsiders. In this paper, we identified different profiles (leaders, brokers, outsiders) based on personal attributes and positions in networks. Yet our approach remains intuitional at this stage. Can we rely on automorphic equivalence to model persons’ positions in a more systematic way? Can we apply blockmodeling procedures to compare more systematically roles patterns across space and time? Although the last set of networks we proposed (roles-roles) may be a first step in this direction, it remains but an intuitive attempt to blockmodeling. Can we model the evolution of the Rotary network on more rational grounds?
Information extraction was a time-consuming and cumbersome process. Role detection and classification depended on subjective choices. Can we tap automatic methods for mining, extracting and detecting roles in newspapers, instead of the manual extraction and a priori categorization of roles we relied on in this research? This would allow us to work on role occurrences directly and to build network of words (i.e. the language of historical actors) instead of analytical concepts (the language of historians) that inevitably introduce biases in network exploration and interpretation. At this stage, however, Named Entity Recognition (NER) and other Natural Language Processing (NLP) techniques do not provide satisfactory results when applied to the English-language press in China. The output they produce still requires a fastidious operation of data cleaning. We might be more successful with Chinese-language newspapers, but this means an entirely new research. Applying NLP-SNA techniques to the study of translation and circulation of role categories and social behaviors across languages and cultures is certainly a promising field of research in the future.
Ultimately, while historical newspapers offer valuable sources for relational data, there may be considerable variations in the quantity and quality information available from one event to the other, across newspapers titles and over time. This have strong implications on the networks built upon such data and the observations we derive from them, especially for researchers pursuing a longitudinal perspective. Can we find a way to rectify the negative effects of data discrepancies in network analysis? Can we harness random techniques to simulate “perfect” networks – i.e. built on ideally complete data – that would enable the testing of historical hypothesis and facilitate comparisons across space and time?
To be continued…