April 11 - 13. Prominent scholars from Groningen held master-classes in SESL
On April 11-13 intensive seminars on network analysis were held in the Laboratory. They were held by invited specialists in this field from the University of Groningen.
What did we need these seminars for? In what areas can network analysis be used? Cristian Steglich started his lecture with answering these questions. He told that this method of data analysis differs much from the classical statistics. The basic principle of statistics in to collect data separately which doesn't happen often in real life. Network statistics takes a network as a core element and studies interconnections of the data. Later on these links are visualised and modelled for interpretation and analysis.
As Cristian Steglich noted, “You don’t need network analysis if there are no interconnections". These interconnections can be seen in the following spheres of social life:
- Administration. Political networks help to understand how states cooperate with each other.
- Trade. Market networks show direction of supplies and relations of traders and customers.
- Organizations. Analysis of formal and informal relations between company employees, especially among management, can be fruitful for revealing corruption, monopolies and reasons of company’s success or failure.
- Primary social organizations. These studies help to analyze kin, friendship and to explain phenomena like social capital in terms of network analysis.
The classes were very intensive. During theoretical part the participants studied the history of network analysis and its key terms. Network analysis was detached from social psychology. Its appearing was connected with group behavior studies and further development resulted from graph theory and practical anthropology advance. Colleagues from Greningen also reminded the basic terms of network analysis to the audience, such as types of networks and actors, reciprocity, density etc.
The participants studied some new methods of data mining and typical problems that happen during this process. One of these problems is defining network boundaries. Where are they? Where does the network finish? The second problem is sensibility to missing values which can spoil the results for the whole network if, for example, a student was absent during the survey. If this student is connected to many people, the network will be irrelevant. The third problem is that people systematically forget certain links and often do not name the closest people.
It was interesting to learn about reciprocity in networks of various types. Its plausibility is higher in egalitarian networks (trust, friendship, communication, etc.) then in hierarchical networks that show distribution of power or chains of supply. Speaking about transitivity, there is certain dependence: plausibility of link between two actors can depend on connection of these two with some third vertex. This thesis helps to measure popularity or social isolation of people. One more interesting thing is “Mathew effect” when the rich get richer.
The classes helped to answer certain questions:
- How can network or actor properties be described by network analysis?
- How can we describe and model dependence between network structure and actor properties?
- How can actors and its properties be compared?
- How do networks develop and influence on each other?
During the practical part of the seminar the participants carried out laboratory work using R. Special attention was given modeling and studying probability models p1 and p2. Probability model p1 shows possibility of network genesis; it was created in 1980-s by Holland and Leinhardt. Probability model p2 considers original independence of dyads; in this model less number of parameters can be used.
It was very useful that the participants managed to discuss their own projects that suppose network analysis with the colleagues from Groningen. Read about this part of the seminar in the next article.
By Ksenia Medvedeva