Working definitions
Goals of social network analysis
Three examples of SNA:
Social exclusion from financial services
Binge drinking behaviours
Technological webs vs. social communities
Conclusion & implications
Cf.
Hayek: social structure emerges from the laws (“nomos") governing the relations between individuals
Potts/Dopfer: cultural rules can be economic and noneconomic, are carried by individuals, diffuse as meso trajectories
Determine most likely structure of relevant social network
Why?
Difference in structure → difference in dynamics
Can tell us about vulnerability to epidemics, speed of diffusion, fragility
‘Friends of friends’
Individuals typically influenced only by close associates
In general, no single individual will influence everybody
Ideas need to be quite contagious to spread
Most vulnerable to ‘epidemics’
Why? A small number of highly connected individuals
Small probability but high impact
No critical level of contagiousness needed for idea to spread
Usually SNA requires vast data sets on every node and connection – to map the network in detail
But Ormerod (2007)* provides an alternative technique to infer the network structure using very small data sets
Relatively cheap (e.g. just 388 respondents provides representation for entire UK population)
Use modelling to assess likely diffusion patterns
Almost one in 10 adults in Britain do not use mainstream financial services. Most of these are not in paid employment.
However, most people without paid work have bank accounts.
Two hypotheses have been put forward to account for the behaviour of the minority without accounts:
reluctance by financial institutions to serve low-income customers; and,
information failure on the part of non-consumers.
First, we can show that having an account is highly correlated with having more f&f who also have accounts:
Use of accounts by friends and family | Proportion of pepole who have accounts | Proportion of pepole who do not have accounts |
---|---|---|
All or most have accounts | 87 | 38 |
Some have accounts | 6 | 26 |
Few or none have accounts | 2 | 14 |
Don't know | 6 | 21 |
TOTAL | 100 | 100 |
Source: ONS Omnibus Survey March/April 2000
Logistic regression shows that social network info is a more powerful predictor of non-usage than all the usualattributes: age, gender, ethnic origin, housing tenure, employment status, income or family circumstances
In many complex social and economic contexts, decision makers often pay attention to each other
use the behaviour of others as a decision rule when choosing between alternative courses of action
Why?
limited information about the problem itself and/or
limited ability to process even the information that is available
The decision to use a back account appears to be this type of problem
Still leaves the question:
What type of social network best describes individual’s decision to use a bank account?
Using survey data from the ONS and network modelling, we have shown that a small world best describes the network in this case.
This supports the information failure hypothesis.
Data on alcohol consumption is fraught with measurement problems
Some series suggest a sharp rise in binge drinking, especially amongst the young, some suggest a drop since 2002
Strong media/public/policy consensus that ‘booze Britain’ is a real and worsening phenomenon
Many traditional econometric studies of alcohol consumption over time and relating changes to factors such as disposable income, price and advertising
All are inconclusive and cannot clearly separate correlation from causation
What if binge drinking is a social network phenomenon?
Precedence: NEJM (2007) quant analysis of the the person-to-person spread of obesity as a possible factor contributing to the obesity epidemic
Using data on 12,000 people monitored from 1971 – 2003
Found that social influence is very powerful: the chance of any individual being obese increases by 57% if s/he has a friend who becomes obese
We are currently applying this approach to answering the question: is binge drinking a social network phenomenon?
Simple survey of ~500 young people
Their own drinking behaviours
Drinking behaviours of family, friends and colleagues
We are currently assessing the most likely network structure: expect small-world
Potentially major policy implications for UK and Europe
Most physical/technology-based nets are scale-free e.g. internet, power grids, air routes
However they differ from social networks in one important respect:
highly connected nodes tend to be connected to less well-connected nodes
By contrast, in social networks, highly connected nodes tend to be connected to other highly connected nodes
This gives rise to community structure (assortative networks)
SNA a powerful analytical technique for uncovering structure emerging from many types of rules/ behaviours
Usually possible to infer the network structure with small amount of data:
Info on the overall frequency of the rule in the population (e.g. what proportion of people UK-wide have a bank account)
Sample data on rule usage by people in the social network
Some evidence on the clustering coefficient (for people-based networks can generally safely assume it is considerably greater than zero)
Is using SNA a way to unpack/formalise phenomena usually referred to as ‘cultural’?
the culture of not using bank accounts
the culture of booze Britain
the culture of collaboration within an organisation
At the very least it shows that structure matters for emergent properties that evolve over time
And provides a very accessible technique when small survey samples and network modelling are combined