Mitchell Centre Seminar Series
| Dates: | 17 February 2016 |
|---|
| Times: | 16:00 - 17:30 |
|---|
| What is it: | Seminar |
|---|
| Organiser: | School of Social Sciences |
|---|
|
|
Bruce Edmonds, Manchester Metropolitan University
How can we check Social Network models and measures? Agent-base modelling as the next level test after face validity.
All social network analysis of observed systems rely on assumptions, for example: how a link is defined is the right one, how the resulting network is analysed actually corresponds with our conclusions about it, etc. In other words the representation+analysis is a *model* of what we observe. Any model is fallible and thus needs independent validation, but this is rarely done in social network analysis due to the cost. Indeed, the only check is often that of face validity by the same person who collected the data and analysed it!
This lack of established validity is somewhat hidden by the divide within the field of social networks between the "formalists" who prove abstract properties of networks and those who apply its techniques to observed cases (who I will call "practitioners"). The formalists might propose SN measures and prove their properties, but do not say anything about their applicability to any observed system. The practitioners often proceed as if the measures will "work" on their networks - e.g. that a measure of centrality will tend to highlight the most influential actors.
However, agent-based models (ABM) might offer a potential solution to this. If a measure (or other SN technique) does not work with a plausible ABM of the phenomena (where we can actually check this), then we certainly can not rely on it for a similar model of observed phenomena. Some results and examples of this are given. Rather, it might be that SNA might be more reliable as a secondary analysis -- a model of a complex ABM of observed phenomena
Travel and Contact Information
Find event
G6
Humanities Bridgeford Street
Manchester
