Social network analysis – appreciation and health warning


Its very tempting to think there is a perfect match between social networks and social network analysis – and its true – there is a lot we can do with this to examine online social networks. However in doing this its important to understand what the limits of what we can achieve with it without bringing in other research methods.

Social network analysis is simply a way of visually and statistically mapping the relationships and patterns of behaviours between the actors in an specific social grouping – it could be familial, social, professional or geographical. Its not a new technique – its been around since the 1930s and has been used to map historical populations such as the Medici’s (Padgett and Ansell 1993), social protest movements such as the the Freedom Summer movement in 1964 (MacAdam 1986, 1988) and to map more contemporary university environments (I have a brilliant reference for this and now can’t find it – damn – will be back to this). Its been used to track the progress of epidemics and its been used by Barry Wellman to look at how communities interact (Hampton, K. N., & Wellman, B., 1999). If you want a quick intro to the ideas here then have a look at this Journal of Computer Mediated Communication Journal article that has the key ideas in it.

In many ways social network analysis is about the whole being greater than the sum of the parts with the interactions and patterns formed from these interactions being as significant as the individuals who are interacting. With it sociologists moved beyond looking at individuals and who they are and started considering what they are actually doing.

Given the fact that social networking is the definitive activity of the social web it is not surprising that social network analysis is a primary tool for their exploration and understanding. However to make this analysis useful we really need to understand the social implications of the links that are created when you friend or follow someone. ‘Follow’ can mean a lot of different things in the twitterverse. Initially it was a politeness – she’s followed me, I should follow her back. Once the spammers turned up then we all got more careful about this and so ‘follow’ becomes more active rather than cultural – who you follow says something about you and you may be consciously aware of this.

That’s not to say that its not useful to map a local twitter network for example – I’ve been working on this this week for a client and once you have an idea of who the principal civic actors are then analysis of their online relationships via twitter can give you a sense of how/if they are connected. Its possible to be more sophisticated about this purely data led exploration – for example you can look at @mentions which can give a good sense of proximity if they are reciprocal and also retweets which can be a measure of reach. But in terms of analysis of relationships its not accurate to say that you can map a social network by this method – we just don’t know enough about the motivations and meaning of any of the measurable actions.

Its also worth noting the difficulty of sampling real time networks such as twitter – its a moving target and if you are looking at interactions such as @mentions rather than just follow/followed relationships then you need to make some sensible decisions as to when not just what you will be sampling.

Its also worth noting that you are strictly limited to ego-centric network analysis rather than looking at whole networks as lack of consistency about association between online and offline identities and also the unbounded nature of our online relationships means that we do not fit out interactions within neat and complete networks such a single organisations or associations. You have to take into account the ability of online social networks to connect you beyond your usual social sphere – to access expertise or interest, support or scrutiny from further away – both socially and perhaps geographically – and it seems that to use social network analysis with online networks you need to understand the underlying social structures first. This spread of connection has been one of the principal criticisms of online networks:

“…some analysts have feared that email, the Internet, and other reduced-cues CMCs are unable to sustain broadly-based, multiplex relations (see the review in [Wellman et al., 1996]; [Garton & Wellman, 1995]). These fears are extended by the boutique approach to online offerings which fosters a specialization of ties within any one of thousands of topic-oriented news groups ([Kling, 1995]; [Kollock & Smith, 1996]). However, this tendency toward specialization is counter-balanced by the ease of forwarding online communication to multiple others. Through personal distribution lists Internet participants can sustain broad, multiplex, supportive relationships ([Wellman & Gulia, 1997]; [Wellman, 1997]). As yet, there has been little research into the extent to which specialized, online, single relations grow into multiplex ties over time.” (Wellman, 1997)

However – the fact that our networks online are in most cases diffuse and unbounded means we need can’t define a network and then analyse the relationships within it in a meaningful way unless we are given access to the complete data set from a site like Facebook (I can really see that request going down well) and then find a really really big computer to deal with it. Instead we rely on an ego-centric approach which unravels relationships based on a single individual or core of individuals – this is in fact called the snowball method – which seems rather apt.

So, we can use twitter (which is more open to interrogation than Facebook and so an easier starting point) in order to do some initial analysis and to start that snowball rolling but to put more detail on this we would need to use other research techniques.

This is more consistent with the literature – more detailed Social Network Analysis data is usually collected via questionnaire and interview techniques rather than analysis of online networks – or rather I have not found any case study examples based on analysis of purely online environments that have not been supplemented by these methods even where they have carried out details log and traffic analysis. There are multiple case studies which have looked at email traffic for example but these will be cross referenced with demographic databases (not surprisingly you can find a number of studies that look at the social networks within academic institutions).

Surveys or interviews usually take the form of either name generators – which ask you who you know or name interpretors which ask you to define your relationships with the people in a described group. Where we start with someone’s twitter followers/followed it makes sense to follow this up with a name interpretation questionnaire which also has the option to add in offline contacts with the same type of relationships. If I want to use this technique in order to explore the relationships that we find between civic creators – something that would be extremely useful in terms of understanding the local civic online space then I will need to find some way of describing the relationships that we might find. Given that I will have chance to do this with a couple of research sites I will be exploring this in preliminary interviews with some of the principal actors that we have found through the research and then asking them to define some of these relationships for themselves. I prefer to do this rather than try to impose relationship descriptions because relationships – as we all know – are complicated:

“Relations (sometimes called strands) are characterized by content, direction and strength” (Wellman et al 1997)

We are trying to understand firstly how people relate to each other and then the degree to which they do so – their centrality. Centrality is a measure of how many connections an individual has but is also an important measure with respect to the whole network. Its an obvious fact that denser networks are more resilient to have an individual member removed but may also be more resistant to new people or ideas. Centrality can also help us to understand where individuals are critical to holding networks together and where their removal might form a cut-point in the network.

Ties are, in this context, either strong or weak. Of course the cut-point for a network may also be considered to be a bridge – or tie – between two different networks. Either way they are useful people to identify. The seminal paper on this “The Strength of Weak Ties” (Granovetter, 1973) explores the idea that weak ties are in fact more effective for passing information and learning new things. He describes the strength as ties like this:

“Most intuitive notions of the “strength” of an interpersonal tie should be satisfied by the following definition: the strength of a tie is a (probably linear) combination of the amount of time, the emotional intensity, the intimacy (mutual confiding), and the reciprocal services which characterize the tie.” (Granovetter, 1973)

So though it is important to understand the strength of ties within a network I am really not sure what we will find when we explore this within a group of civic creators in a locality – another good reason to do the research.

Unless you are going to take a structural deterministic view of things and say that the networks are causing the behaviours of individuals rather than just describing them (which I’m not) then social network analysis is primarily a tool for exploration and discovery – a way of seeing groups of people in new ways in terms of the way in which they interact and potenitally influence each other.

This brings me my final point on the limitations of social network analysis with respect to civic communities – privacy and identity. We already know that many people prefer to keep their identity anonymous when talking about political issues (have a read of Michelle Ide Smith’s findings which are relevant at the hyperlocal level) which means that we may only be able to reach these people online and they may not be willing to share more information about who they are connected to because to of the inferences that you can make from this knowledge at a local level.

I think one other consideration also has to be to what extent your survey sample understands the publicity of their actions or whether they will find an analysis of their twitter interactions to be intrusive. This is another issue which I will be approaching with caution with some initial interviews as we are planning on gathering a group of civic creators whom we have isolated through an online research process and will see whether they feel surveilled or appreciated.

Social network analysis is a visual and accessible technique that can provide you with a quick snapshot of activity and connections for a network. It can expose connections you were not aware of and make clear the importance of individuals who you may not have been aware of. However, its unlikely that its going to do this if you rely just on the output data sets from online social networks themselves as we do not have enough clarity and agreement about what the relationships captured with the follow/followed actions actually are. Twitter (and other sites) are an excellent starting point for a snowball approach to data collection but to get robust data you need to ensure that you carry out follow up surveys and where possible interviews.

As a final point – its interesting to speculate what a network like Path might mean in this context. Path limits your network to 50 people and is specifically targeted at sharing things with friends and family rather than creating a wider network (good Wired article on this here). If they manage to create critical mass then it will be interesting to compare the data from Path with more conventionally gathered survey data – and to see if your what your network looks like if take physical world limitations of scale and apply them online – but this falls clearly into the category of things to think about post PHD.

PS this finally gets to the bottom of exactly why I find Twinfluence and its friends a little like social media snake oil…..very satisfying

3 comments
  1. curiouscatherine

    April 26, 2011 at 10:08 am

    Thanks Tony…

    Have been using NodeXL for the @mentions but your graph looks better on this and I really like the hashtag map. Assume the links are where people are following each other?

    Am currently looks for ways that you can follow a RT as through a social network – I want to know who RT’s who as its a better indicator of influence than just following- don’t suppose you have anything on that?

    Thank you!

    C

    Reply
  2. Tony Hirst

    April 26, 2011 at 4:09 pm

    @Catherine

    I use Gephi mostly, but as a desktop app; the Protovis javascript library meant I could do things on the fly in the browser.

    I haven’t done RTs but I have started to looked at several variants of link tracking, though there are quite a few improvements I need to make as to how I present the display (in particular distinguishing between link tweeters and audience):

    a) using BackType to see who has tweeted a particular link e.g. http://blog.ouseful.info/2011/02/03/visualising-ad-hoc-tweeted-link-communities-via-backtype/

    b) using bit.ly and BackType to see who has been tweeting links shared by a particular individual e.g. http://www.flickr.com/photos/psychemedia/5437539168/

    c) using delicious e.g. http://blog.ouseful.info/2011/01/08/visualising-delicious-tag-communities-using-gephi/

    So rather than at RT, per se, I have started to try to track folk based on hashtag use, or the tweeting of a particular link. I guess that Twitter-RTs carry the original tweet ID as metadata (and maybe replies do to?) but old-style RT @x: may not, and would require some textual analysis to decide whether two tweets are the same.

    The above recipes use personal API keys, so I don’t really want to post them as View Source-apps webpage apps; I’ve more or less sussed doing simple things in Google App engine, though, so given a couple more playtimes I may well soon be able to post some link tracking follower graphs as a live webapp… err… maybe!;-)

    If you have a specific project/use case in mind, maybe we could bounce a few more ideas round?

    Reply

Leave a Reply

Your email address will not be published. Required fields are marked *