What might formal modelling tell us about the spread of radical behaviours?
A brief reflection on: Törnberg, A., & Törnberg, P. (2017). Modelling free social spaces and the diffusion of social mobilization. Social Movement Studies, 16(2), 182–202.
Social Movement Studies have recently published an interesting article making use of formal computational modeling to examine the ways in which ‘free social spaces’ contribute to diffusion processes. This blog post begins with a brief review of the paper and then considers ways in which its findings might be extended to help make sense of all the talk of ‘echo chambers’ that has emerged since the election of Donald Trump. I close with a suggestion that future computational modeling might fruitfully focus on competitive diffusion processes.
‘Modelling free social spaces and the diffusion of social mobilization’ by Anton Törnberg and Petter Törnberg is an intruiging and original piece, one of very few attempts to introduce this methodology to social movement scholarship. [1. Although Pamela Oliver and colleagues have done some work here, e.g. Oliver & Myers (2002) The Coevolution of Social Movements in Mobilization 8(1): 1-24.] The approach’s deliberately simplistic assumptions and mathematical abstractions might seem like a hard sell, particularly for qualitatively inclined sociologists. But Törnberg & Törnberg are very sensitive to the need to root their analysis in accounts of diffusion developed in a literature making use of a range of methods as well as the importance of clearly acknowledging the limits of possibility with their chosen approach. So it seems only fair to suspend my own disbelief about the potential for modelling such complex and context-laden social interactions and see where the argument leads.
The authors note that the literature on free spaces (with multiple kinds of actual empirical referents) tend to highlight their potential contribution to episodes of heightened mobilization. They work with two key elements: in network terms, such spaces are clusters of people whose ties are dense relative to the wider society in which they are located; and members of the space are in some common way deviant from members of the wider society. Starting with a network that contains a single cluster and a set of less connected nodes, then, the model is used to examine the way that something, once introduced to the network, will diffuse across it. Ultimately the authors are interested in the mobilization of protest, so what is diffused might be a willingness to participate in a particular kind of protest. (Direct action against the summits of international financial institutions, say.) The ‘deviance’ of the cluster is modelled in terms of their bias toward the diffused object – a higher bias would mean that individuals, once exposed to that object, are more likely to pass it on. Slightly more technically: diffusion is seen as complex contagion, wherein any individual node would need more than one neighbours to be activated before it is also activated by the diffused behaviour or preference. If a node is biased toward the diffused behaviour then the threshold for how many neighbouring nodes are required for diffusion is lower than the average for the societal network as a whole. In real terms we could imagine a group of squatters with autonomous politics finding out for the first time about alter-globalization protests. In their dense network, if a few people take part in the protests they are very likely to spread that potential for mobilization to others, not only because they are densely connected but also because others are easily persuaded (in this case because of the match between direct action against neoliberal instiutions and autonomous politics).
So far, we’ve not got much further than the basic insight that pre-existing social networks can, under certain circumstances, be mobilised en masse. But what this paper is interested in is the society-level effect: does the uptake of the diffused practice then spread out of the cluster to wider society? This is where the modelling is highly revealing. In summary, it reveals a two-step diffusion process wherein dense clusters contribute to a local mobilization, which is necessary before a second phase that activates long-range ties beyond the cluster for more general societal moblization. However, the optimal level of clusterness for successful social mobilization depends in part on the degrees of bias towards the diffused practice that exist in the network. Where the cluster contains high levels of bias (which is used to indicate more deviant or radical beliefs) then high levels of density are actually counter-productive in terms of wider social mobilization. (This chimes with, but is much more specific than, the famous ‘strength of weak ties’ idea from Granovetter.) In biased clusters it is easy to get the first phase local moblization process off the ground, even without a very high density of ties, the challenge is in spreading it into wider society. If, on the other hand, the cluster is only a little more biased compared to the wider society (and so we’re thinking about diffusion of a less radical practice) then high density of connections is required to get that first phase started; once it has started, however, there is a good chance of spreading mobilization across the wider society.
“clusters with a high bias need more external connections (less clusterness) to be efficient in spreading activation. Thus, in order to diffuse more effectively, radical free social spaces might gain in trying to increase the number of (external) connections connecting the group with the rest of the society. The same goes the other way around; free social spaces that are not radically deviant from overall society might gain in increasing the ratio of internal ties and to build strong ties within the group in order to facilitate the global diffusion of mobilization.” (199)
The analysis thus connects to a frustrating truth for activists in that, often, the more ‘radical’ the group, the harder they find it to be genuinely open. No matter how ‘inclusive’ your ideology, if your practices include four-hour meetings in unheated squats with skipped food for refreshments, it really is harder to get ‘ordinary people’ into your circle. Yet it is precisely such groups that need the most connections beyond their own clusters if they are to contribute to the mass spread of their preferred beliefs or actions. On the other side of the coin, consider a more moderate practice like ethical consumerism. You don’t have to be terribly ‘deviant’ in a capitalist democracy to believe that consumer choice might be utilised to discourage corporations from massive human rights violations. What the modelling suggests is that for that practice to have got off the ground in the first place, ‘moderate’ clusters would need a high density to develop initial mobilization, after which it would have been relatively easy to spread out further. In some ways this opposes the kind of depiction of ‘lifestyle movements’ found in, for instance, Haenfler et al. [2. Haenfler, R., Johnson, B., & Jones, E. (2012). Lifestyle Movements: Exploring the Intersection of Lifestyle and Social Movements. Social Movement Studies, 11(1), 1–20.]
I’m probably already going much further in filling out the abstract results of the formal model than Törnberg & Törnberg would really allow. They are clear that such models don’t provide a proof of such structural conditions, but should be seen rather as a way of thinking through possibilities in a way that extends our intuitive theorising. Any real set of social interactions obviously include much that the abstract model does not, which may well confound the patterns here, or produce different ones. Nevertheless, such thinking does seem to be highly suggestive. To extend even further, then, can Donald Trump’s unexpected election victory be considered in the light of these sorts of insights? To be successful the preference for a US President Trump – once seen as a joke that wouldn’t make it through the primaries – had to diffuse across much of the electorate. Attempts to make sense of this outcome have already covered a lot of ground. Some proposed answers have highlighted ‘echo chamber’ or ‘filter bubble’ effects that allow individuals to tailor their news sources to their own pre-conceived biases, untroubled by pesky alternative views or confounding evidence. (And search engines and social networks are often successful precisely because they are good at doing this filtering for us on the basis of our past behaviour.) The fundamental idea of the echo chamber is that radical ideas get repeated, and perhaps radicalised, within one’s chosen closed group communication. We might see clustered communication as an example of ‘free social spaces’ in Törnberg & Törnberg’s sense. If echo chambers are very closed and very deviant though, this model would suggest that diffusion is unlikely to get far beyond the originating cluster. By implication, there is little risk of such spaces leading to widespread shared behaviour like voting Trump and so echo chambers cannot explain the observed phenomenon. [3. In fact, what echo chambers may be better at explaining, in relation to both Trump and Brexit, is the fact that, stuck in our own echo chambers, the left didn’t see it coming.] Given that wider social mobilization did succeed, then, perhaps what we should be focusing on is the possibility that either: a) the spaces were not that closed (i.e. there were plenty of avenues for this to diffuse into the ‘mainstream’); or, b) that the preference for voting Trump was not particularly ‘deviant’ (i.e. that despite the disgust of liberals there was sufficient ideological connection between Trump and ‘the mainstream’ to mean that the number of connections between the chambers and wider society were not especially important).
Of course, there are plenty of ways in which reality doesn’t match up to any abstract model. Politics and economics matter hugely. Törnberg & Törnberg fully acknowledge this. But, given these kinds of examples can we reflect back on the nature of the modelling process? One point I would highlight is that in the real world there are always multiple messages competing for attention and adherence. As with much of the diffusion literature in general, this paper focusing on diffusion of a singular message or practice, so that agents’ choices are limited (accept or reject). But the world of politics is always competitive: at least vote Trump vs vote Clinton in this example; or, say, orderly marches vs smashing shop windows in the alter-globalization movement. It may be fruitful to think about the multiple messages as being in competition for diffusion. If it is possible for future computational modelling to examine the relationship between communication structures and competitive diffusion processes this might – to paraphrase the authors – lead to ways in which we can extend our insights beyond what our natural intuition allows.
Kevin Gillan
Kevin Gillan is a lecturer in sociology at the University of Manchester and Editor in Chief at the journal Social Movement Studies. He has interests that overlap between social movements scholarship and economic sociology.
0 Comments