That’s what you get, when you ask people to draw their networks, they (most of us) will draw a network that makes them look much better connected than most other people in the network. That does not necessarily show their delusion of grandeur but just reflects the fact that you know much more about your own links than about other people’s links.
We are running into this problem at the moment when interviewing female rice processors in northern Nigeria about their business networks: If you naively looked at these drawn networks, without knowing, which network actor was the interview partner, you would think that these women are the best connected people in the whole of northern Nigeria. So, if we still want to look at the quantitative measures, what do we do?
If we want to compare centrality values within the networks (Who is most central in this specific interview partner’s network?), we compare everyone’s but the interview partner’s value. Not that we would cut them out of the network when calculating (because then we would have an unrealistically disconnected scatter of actors) but we don’t include their extremely high centrality values in our comparison. If we want to compare between different interviews however, it might be very interesting to compare: How high are the centrality values of different interview partners and to ask e.g.: Is there a connection between how connected an interview partner sees herself and how successful she is in her business? For comparing between interviews it makes sense to use the “normalized” centrality values. They basically indicate what percentage of the highest possible centrality the actor has. So a normalized degree centrality of 70% means that the actor is connected to 70% of the actors in this network.
Filed under: technical details |