Panel Random and Digital Networks

The great majority of research studies on social digital networks focuses on interindividual relationships (“friendship” or “followers” networks) and represents them by graphs. On another hand, works about individual internet users’ attributes give a greater place to qualitative approaches on smaller corpus.

This communication shows that, to analyze social digital networks, more precisely Twitter, it’s possible to adopt a quantitative approach, taking account of individual users’ attributes. Panelisation method, applied to the “twittosphère”, let us to reconcile these two exigencies. Analyzing reasoned sample, panelisation allows to qualify Twitter accounts manually (sociodemographical attributes or audience data, etc.) and allows to explore message contents on the long way.

In these conditions, the question is: according to what logic can we sample Twitter? Is a panel randomly selected really representative of the network? Does Law of large Numbers apply to a socio digital network where every individual doesn’t have the same weight neither the same visibility? Actually, on Twitter, accounts don’t have the same visibility neither the same influence. How taking account of these inequalities?

We experiment two sampling logics. The first with a random sample: every account has the same weight, whatever its activity or its influence. From an exhaustive list of the population, we randomly select a panel of 2000 accounts. The second logic takes account of the hierarchical structure of Twitter. The “influence panel” is constituted from a list of accounts classified according an index of influence. We select the 1000 first accounts of the list.