Social Network dynamics shift under stress

Fascinating article in New Scientist, looking at the dynamics of how email network behaviour changes in different conditions:

Ben Collingsworth and Ronaldo Menezes at the Florida Institute of Technology in Melbourne identified key events in Enron's demise, such as the August 2001 resignation of CEO Jeffrey Skilling. They then examined the number of emails sent, and the groups that exchanged the messages, in the period around these events. They did not look at the emails' content.

Menezes says he expected communication networks to change during moments of crisis. Yet the researchers found that the biggest changes actually happened around a month before. For example, the number of active email cliques, defined as groups in which every member has had direct email contact with every other member, jumped from 100 to almost 800 around a month before the December 2001 collapse. Messages were also increasingly exchanged within these groups and not shared with other employees.

Menezes thinks he and Collingsworth may have identified a characteristic change that occurs as stress builds within a company: employees start talking directly to people they feel comfortable with, and stop sharing information more widely.

One of the issues for researchers is getting access to enough emails to study as privacy regulations prevent this, even for academic study. Another study by Yahoo shows:

Duncan Watts of Yahoo Research in New York managed to obtain anonymised university email records - though he is not able to share them with others - and used them to show that instead of varying continuously, [different] individuals' emailing behaviour falls into distinct clusters:

We find that the resulting best-estimate parameter distributions for both data sets are surprisingly similar, indicating that at least some features of communication dynamics generalize beyond specific contexts. We also find that variability of individual behavior over time is significantly less than variability across the population, suggesting that individuals can be classified into persistent "types". We conclude that communication patterns may prove useful as an additional class of attribute data, complementing demographic and network data, for user classification and outlier detection---a point that we illustrate with an interpretable clustering of users based on their inferred model parameters.

We've seen work that shows that a customer's benaviour online can give away the "shadow of future intention to defect" and Telcos have been able to infer a lot from their own users' behaviour. In other words, there is much further to go here.