Archive for the ‘Concepts’ Category

Paper on the micro-side of network analysis

February 10, 2010

Spoken Networks: Analyzing face-to-face conversations and how they shape our social connections.

Tanzeem Choudhury (Dartmouth College)

With the proliferation of sensor-rich mobile devices, it is becoming increasingly easy to collect data that capture the real-world social interactions of entire groups of people. These new data sets provide opportunities to study the social networks of people as they are observed “in the wild.” However, the traditional methods of social network analysis are often inadequate for such behavioral data. Most existing techniques apply only to static, binary data. Social networks derived from behavioral data will almost always be temporal and will often have finer grained observations about interactions as opposed to simple binary indicators. Thus, new techniques are needed that can take into account variable tie intensities and the dynamics of a network as it evolves in time. In this talk, I will provide an overview of the computational framework we have developed for modeling the micro-level dynamics of human interactions as well as the macro-level network structure and its dynamics from local, noisy sensor observations.

Furthermore, by studying the micro and macro levels simultaneously we are able to link dyad-level interaction dynamics (local behavior) to network-level prominence (a global property). I will conclude by providing some specific examples of how the methods we have developed can be applied more broadly to better understand and enhance the lives of people.

(To learn more about Tanzeem visit:


Center for Complex Network Research (CCNR SEMINAR)
Dana Research Center
110 Forsyth Street, 5th Fl. (Large Elevator)
Boston, MA 02115

This blurb courtesy of David Lazer.

Apropos of our discussion about triads …

February 6, 2010

In class we were discussing whether triads were the minimum structures of network interest. Take a look at this post

and the paper it discusses:

Faust, Katherine. 2007. “Very local structure in social networks.”  Pages 209-256 in Sociological Methodology 2007, volume 32 edited by Yu Xie.  Cambridge, MA: Basil Blackwell. [pdf]

Quick overview of network analysis for a reporter

March 4, 2009

Hi, I understand you are looking for a high-level description of UCINET and what it is used for.

Given network data (such as who is kin of whom, who is boss of whom, who talks to whom, how gives money to whom, etc) the program computes a bunch of metrics that illuminate the structural role played by individual nodes in the network (or specific relationships, or entire groups).

So, for example, it can model the expected amount of information that is flowing through the network that will reach a certain node, and model when it will receive it. Typical guiding theoretical principle is that nodes that are more central in the network (in various different senses), have certain advantage over other nodes, which enables us to predict that they will perform better than others, or get more rewards, or will increase in status/power. Etc. At the same time, we can estimate which nodes will do the most damage on removal from a network, in terms of damaging that network’s abilities to transmit information and orders.

 Academic Research. At the most general level, UCINET is used to investigate two kinds of questions: (a) variation in performance/success, and (b) homogeneity in attitudes, beliefs and behaviors.

 Research into the first question can be broadly termed social capital research – it is about the benefits of social ties and social positions and ultimately network structure. It answers the question, what does it get you to be located where you are in a network that has a certain structure. So classical work on how people get ahead, how do you get jobs, who’s got the power, why are some more creative than others, etc are all part of this research stream/. 

Research into the second question can be broadly termed diffusion or peer influence research. It is about how one’s beliefs, attitudes, practices, and so on are shaped by the people (or other entities) we interact with. Classical work on language acquisition, becoming a criminal, fashion and consumer marketing, politics, and so on part of this research stream.

Applied Research. All of the academic work has obvious implications for many applied fields – for example, the diffusion research is key to shaping how pharmaceutical companies identify physician “key opinion leaders” and try to influence the influencers. But I would say that there are three well-developed applied SNA areas: criminology/terrorism, public health, management consulting.

 Criminology/terrorism. A key goal here is breaking up criminal networks – “whack-a-mole” type applications where you try to figure out which are the key players such that removing them from the network (i.e., arresting, shooting, discrediting) would do the most damage to the network’s ability to cause problems for others. In this area, much of the work is directed toward figuring out how to obtain the network data in the first place, since the terrorists won’t fill out surveys. Essentially, the problem they have is way too little good data (e.g., who trusts whom, reliably measures) and way too much bad data (terabytes of data linking people to others in highly circumstantial and unreliable ways).

 Management Consulting. The goals here are usually the opposite the criminology goals. Here you want to strengthen the network and help it accomplish its goals more efficiently. The network analysis done by UCINET can be used in a number of ways. For example, in the case of post-merger integration (PMI), you have two companies merging not only their technologies, but their cultures and their people (and with them, their networks). A network analysis quickly tells you where the networks are integrating and where they are still remaining apart. You can also use UCINET to discover key nodes who (a) should be given a strong stake in the company because they would leave a large hole if they left, and (b) have the “network signatures” of future stars, and should be groomed  for promotion, and (c) maybe bottlenecks because they are so good that too much is getting channeled their way, and the system is become slow and brittle (when he has cold, everything grinds to a halt).

 Health. The goals here a combination of the criminology and management consulting goals. On the epidemiological side, you have a contagious disease that is spreading from person to person. By doing a network analysis, you can figure out which individuals (or collectivities) need to be immunized/quarantined in order to slow the spread of the disease as much as possible. This is the same problem as the stopping of a terrorist network problem. On the other side, there is the patient care side, in which patients get the care they need by having helpful and knowledgeable friends that are either doctors themselves, or are good at asking questions of doctors, or who can provide good referrals, etc. Also patients who are able to connect their various specialists so that they talk to each to each other can achieve much better outcomes for themselves.

Groups versus Networks

January 24, 2009

What’s the difference between a group and a network? There is considerable confusion about this, and the question itself is complicit in the confusion.

Groups and networks are not alternatives to each other. We can point to a big leafy thing in our backyard and ask is that a tree or a bush? The dividing lines between trees and bushes maybe quite blurred, but the question is reasonable.

In contrast, asking whether something is a group or a network is anot a sensible question. A group defines a set of people, and the set of ties among those people is a network. Every group has a network as one of its aspects. So does any collection of people, such as the set of people attending a certain class. 

Furthermore, networks need not be connected. For example, at the beginning of a semester, the people attending a certain class may not have any connections, direct or indirect, with certain other members of the class. Later, these connections may develop. But it is always a network. 

A difference between groups and networks is that a group defines a set of actors, and a set of actors defines a network. 

Evolution of networks. It troubles people that networks can have no ties, or can have disconnected components. But this is important because it allows us to observe network evolution as it really is. Moody has some data showing that different components of a network form before it becomes connected, rather than starting from a core and diffusing outward.