[Research notes] Survival and turnover in ana-mia online networks. An in vivo study of the effects of moral panic surrounding eating disorder websites


Antonio A. Casilli1 2, Fred Pailler2, Paola Tubaro3

1 Télécom ParisTech (France);

2 Edgar Morin Centre, EHESS, Paris (France);

3 University of Greenwich, London (UK) and CNRS, Paris (France)

Moral panic surrounding ana-mia websites

In February 2012 Tumblr, the second most popular microblogging platform after Twitter, announced its decision to remove all contents related to self-harm[1]. These include suicide, mutilation and most prominently thinspiration, the ritualised exchange of images and quotes meant to inspire viewers to be thin. This practice is distinctive of online blogs, groups, and communities known as pro-ana (anorexia nervosa) and pro-mia (bulimia).

Online ana-mia content creators are prevalently teenage and young women living with eating disorders, and secondarily sympathizers (sometimes dubbed wannarexics, i.e. “wannabe anorexics”). In media narratives, they pose as heroic sufferers, and go as far as calling their eating habits a lifestyle “choice” rather than a disease[2]. Our research reveals a more complex picture. Although these websites offer everything from tips on starving and purging to photos of celebrities doctored to look unrealistically thin, they also provide tools of self-help and empowerment for and by persons with eating disorders. Some of them offer online support for sufferers and occasionally accompany them toward treatment and recovery.

Following Tumblr’s example, one month later the increasingly popular photo-sharing site Pinterest also disallowed thinspiration[3]. Some commentators saluted these decisions as righteous limitations of harmful contents, although arguably also driven by a concern to avoid liability. Here is how an observer related her shock at the sight of thinspiration images on Pinterest at about the time the ban came into force:

Thinspiration on Pinterest ranges from photos of stereotypically attractive women, quotes about how good it feels to work out, or a combination of both.  I found one quote that says, “Nothing tastes as good as skinny feels.”  Another one, with a picture of a thin, almost naked woman says, “What you eat in private, you wear in public.”  Finally, I saw a picture with another barely dressed, thin woman that says, “It takes 4 weeks for you to notice your body changing, 8 weeks for your friends and 12 for the rest of the world.  Don’t quit.”  You cannot log onto Pinterest with out seeing these types of thinspiration show up. Not only are these women posting thinspiration to shame their own inadequacies in their bodies but also with Pinterest, these images get repined [i.e. forwarded] for all of their followers to see so they know too that their bodies are inadequate.[4]

Negative reactions to such controversial images and comments are not new, though; neither is the decision of web service providers to prohibit them. Ana-mia websites have existed for over a decade[5] and have always been under strong pressure. What are, then, the reasons that explain their resilience? And is a ban an effective solution? Following in the tradition of Edgar Morin’s“sociology of the present”[6], our paper proposes an in vivo analysis of survival and turnover social mechanisms in eating-disorders online communities and advances policy recommendations.

How effective is the ban?

AOL and Yahoo were the first to ban ana-mia websites as far back as 2001-2[7], when the movement was still essentially confined to the USA. But the online ana-mia community did not vanish: a first, simple indicator of its resilience is given by Google search trends for ana-mia queries in the following years. Figure 1 represents the world-wide evolution of the search volume index in 2005-2009, showing peaks of interest for ana-mia news stories among the general public.

Figure 1 – Evolution of search volume index for ana-mia-related queries (2005-2009). Source: Google Trends.

These figures have limitations, in that they count only searches and do not reveal the actual number of eating disorder blogs and websites (to say nothing of groups and accounts on Facebook and Twitter, largely untapped by standard web search). But they suffice to prove that the first wave of censorship in the early 2000s failed to reduce interest in the phenomenon. What’s more, ana-mia websites spread to other countries afterwards: initially observed almost exclusively in the USA, they now exist in multiple locations and languages. Figure 2 shows searches for ana-mia contents in the second half of the decade in France and the UK, two countries where public attention to the phenomenon in recent years prompted proposals of restrictive legislations (2008)[8]. Providers have been increasingly reluctant to provide domain hosting, and blogging platforms have sometimes shut down individual pages. And yet, the recent moral panic around Tumblr and Pinterestr demonstrates that ana-mia blogs, forums and online groups have not disappeared.

Figure 2 – Search volume index for “pro-ana” queries, UK and France (2004-2010). Source: Google Trends.

The structure of ana-mia online networks

How large is the ana-mia webosphere today, and how has it managed to survive for so long, despite actual or perceived restrictions? To answer these questions, we have built a web corpus of the French ana-mia webosphere between 2010 and 2012 using Navicrawler, a specialized web-mining tool, and Gephi, a software package for exploratory data analysis and visualisation. It is not an exhaustive representation: these tools can capture blogs, forums and web sites, but not social networking services such as Tumblr, Pinterest or Facebook. Yet they suffice to reveal key features of this otherwise invisible portion of the Internet. Figure 3 is a snapshot of it in March 2010 and March 2012. Nodes represent web pages and edges represent links between them. The structure of links between pages provides a global picture of the communication patterns within this part of the Internet, and indicates how a user (or a regulator or even a censor) can navigate it to discover ana-mia contents by browsing, starting from any one of the websites in the map.

Figure 3 – The French ana-mia webosphere in 2010 and 2012. Node size represents the number of incoming and outgoing links, in three categories (small, medium and large). Colour represents beetweeness centrality, also in three categories (red = low, purple = medium, blue = high). Source: Authors’ elaboration.

At first sight, the two networks look very similar. Indeed there are about the same number of sites at the two dates, 559 in 2010 and 593 in 2012, indicating that the size of the network has not shrunk despite all pressures to ban ana-mia contents. While as discussed, these figures have no pretension of exhaustiveness, they nevertheless provide evidence that use of the same exploratory tool at two different dates yields about the same global result, which is a clear sign of stability.

Notice, also, that both networks are composed of sizable cliques, discernible at the top and at the bottom of the graphs. This is a common characteristic of web-based and other large social networks: they consist in tightly knit sub-groups with few connections to one another. Technically, one would say that network “density”, or number of observed links relative to number of potential links, is globally low and locally high. Here, the observed clusters correspond to popular blog platforms. Within each of them, smaller blogs rally around a few “hubs” – individual blogs represented as nodes of larger size in the graph. They are mostly “repository” websites which gather, systematize and re-diffuse informational resources through their links. They are most commonly found in the top clique.


Survival and turnover

Yet this resilience is unevenly distributed within the network: not all blogs continue to exist. A closer look at the data reveals a turnover of about 50%, with only 296 blogs surviving from 2010 to 2012. The community remains alive, but its endurance is only due to the surviving capacity of these long-lasting blogs and the continuous renewal of the ephemeral ones around them.

Figure 4 – The top clique of the network in 2012. White nodes represent surviving websites (already observed in 2010), and blue nodes represent new websites (created after 2010). Node size depends on number of links, both incoming and outgoing. Source: Authors’ elaboration.

What are, then, the strategies of the surviving blogs? They do not dissimulate proscribed keywords: indeed their names are often explicitly pro-eating disorders. The structure of the network suggests that their resilience involves more sophisticated strategies. One indicator is the extent to which nodes are bundled in local interconnected neighbourhoods, or “clustering coefficient”, which can be interpreted as a local measure of density around a single node. On average, the clustering coefficient is higher for surviving nodes than for all other nodes, and is particularly high in the clique at the top of the graph (Figure 4), which is the densest and also the one that includes the largest number of surviving blogs (57%). Survivors are embedded in local neighbourhoods, in dense clusters that become almost impenetrable, and therefore more difficult for external actors to locate. They are better at shielding themselves through a net of interwoven blogs around them.

Other network properties suggest that this uneven density is part of a more general strategy of survivors to get around (actual or perceived) restrictions. First, Figure 3 shows that there are fewer links between the top and the bottom cliques in 2012, relative to 2010. The lower the number of bridging links, the higher the likelihood that the network may break up into separate pieces: it would suffice to remove these few links, or to close the (equally non numerous) websites containing them. Second, there is an increase in the maximum number of steps, or length of the chain of web links that users must click to reach any one site from any other within the network (“diameter” in network analysis terminology). This means that some nodes are somehow pushed to the side, becoming less well connected to the others over time. Third, between 2010 and 2012 nodes lose some of their capacity to act as intermediaries (brokers) between other nodes, that is to connect others that would otherwise be disjoint, as shown by a slight average fall in another measure called “betweenness centrality”. What is the position of surviving nodes in this respect? They have above-average brokerage capacity (technically, higher betweenness centrality) at both dates, due to their role as hubs that gather and redistribute information through their links; yet they do so only locally, within cliques, not between them. They allow global connectedness to diminish, and contribute to reinforcing the crowding together of nodes in local clusters only.

2010 2012
Network density 0.012 0.011
Average clustering coefficient 0.197(Survivors: 0.216) 0.164(Survivors: 0.187)
Average betweenness centrality (normalised) 0.19(Survivors: 0.26) 0.10(Survivors: 0.13)
Diameter 14 22

Table 1: key metrics describing the French ana-mia webosphere (2010-2012). Source: Authors’ elaboration.

Why this is bad news for public health policy-makers

Above, we have reviewed cases in which restrictions imposed by one service or country did not succeed in wiping out the ana-mia phenomenon. Here, we see that even when there is no explicit ban, but only the anticipation of it, it is enough to induce self-protective reactions and reshape the graph. Potential bans are just as effective as actual bans -but not always in the most health-promoting way.

By inducing blogs to converge into one of the bigger clusters, pressure from the outside encourages the formation of densely-knit, almost impenetrable ana-mia cliques. This favours bonding and information redundancy – meaning that ana-mia bloggers will tend to exchange messages, links and images among themselves and to exclude other information sources.

An information and awareness campaign on the risks of extreme fasting or exercise is now less likely to reach out to ana-mia bloggers. If in 2010, such a campaign would have targeted the websites in the middle of the graph so that they relay the message to the margins, in 2012 the middle is virtually deserted, and the margins are located further away. The only option is to try and cut through the dense cliques and hope health messages may somehow reach the key information hubs. Yet the chances of success are significantly lower, as these hubs ensure their survival by hiding, so to speak, beneath a net of increasingly interwoven ephemeral blogs; and they are less well positioned as brokers who can re-transmit the message to large audiences.

These results cast serious doubts on the effectiveness of bans. Ana-mia bloggers react to actual restrictions by moving onto other platforms or to other countries, and anticipate potential restrictions by reshaping their social network in denser and less interconnected clusters. Those who suffer the consequences are healthcare professionals, public decision-makers, families and charities for persons with eating disorders. They know how hard it is to reach out to ana-mia online communities, and they know that it will be increasingly hard if these communities become more suspicious, secluded and inward-oriented.

[The authors would like to thank the members of the ANAMIA research team: Claude Fischler, Christy Shields, Débora de Carvalho Pereira (Centre Edgar Morin, EHESS, Paris); Lise Mounier, Sylvan Lemaire, Manuel Boutet, Pedro Araya (Centre Maurice Halbwachs, CNRS, Paris); Estelle Masson, Christèle Fraïssé, Sandrine Bubendorf (Centre de Recherches en Psychologie, Cognition et Communication, Université de Bretagne Occidentale, Brest); Pierre-Antoine Chardel, Patrick Maigron, Claire Strugala (Groupe de recherche Éthique, Technologies, Organisation, Société, Institut Mines-Télécom, Evry); Juliette Rouchier (Groupement de Recherche en Économie Quantitative d’Aix-Marseille, Université Aix-Marseille 2, Marseille). The research project is funded by the French Agency for National Research (ANR) under grant agreement n. ANR-09-ALIA-001.]

[2] Antonio A. Casilli. (2010) Les liaisons numériques: Vers une nouvelle sociabilité?. Paris : Seuil.

[4] Alexa Megna. (2012) Pinterest’s Thinspiration Problem. Sociology in Focus, April 11, <http://www.sociologyinfocus.com/2012/04/11/the-pinterest-problem/&gt; [last accessed : July 19, 2012]

[5] Antonio A Casilli, Paola Tubaro and Pedro Araya. (2012) Ten Years of Ana: Lessons from a transdisciplinary body of literature on online pro-eating disorder websites. Social Science Information. 51 (1): 120-139.

[6] Cf. Bernard Paillard (2008) La sociologie du présent. Communications, 82 (1): 11-48.

[7] Catherine Holahan. (2001) Yahoo Removes Pro-Eating-Disorder Internet Sites. Boston Globe – Boston, Mass, Aug 4: A.2.

[8] (2007) Proposition de Loi n° 3481 de Monsieur François Vannson tendant à interdire les sites Internet faisant l’apologie de l’anorexie, Assemblée nationale, February 26; (2008) Early day motion 659: Anorexia Web Sites, Primary sponsor: Bob Spink, UK Parliament, February 03.


5 thoughts on “[Research notes] Survival and turnover in ana-mia online networks. An in vivo study of the effects of moral panic surrounding eating disorder websites

  1. Starting from the specificities of the phenomenon “ana-mia” (well described in the first part), this article approaches the question of the evaluation of this phenomenon and of the monitoring of its evolution through communal and informational resources that it mobilizes.

    The social permanency of the phenomenon raises questions about the efficiency of ban against it on the web.

    The proposed method relies on the comparison of two views of the webosphere took at a two years interval under strict experimental conditions. The scientific approach is very clear and well explicited. In particular, web data analysis shows how to build computational representations well adapted and how to transpose their interpretation in the sociological frame specific to this studied phenomenon. From this point of view this article is an excellent reference.

  2. (sorry for the break…)
    To go further (in my opinion), it will be very interesting to develop the characterization and the analysis of mechanisms of “virality” that lead in maintaining the informational structure (system) alive in that case.

  3. The research note “Survival and turnover in ana-mia online networks” focuses on the structure and change of online “ana-mia” communities of eating disorders after bans (or perceived bans) of eating-related self-harm content on popular platforms such as Tumblr or Pinterest. The contribution concludes that restrictions have not been effective as the ana-mia community has not disappeared. To the contrary, a network exploration of the interconnected ana-mia sites shows that the network has become more resilient towards external influences of e.g. healthcare professionals who might want to reach out to the community.

    The note is obviously highly relevant for Just-in-time Sociology, the topic is highly interesting, topical and worth to study. It is well-written and well-researched. I only have a few comments on some (mostly technical) issues which would be good to be clarified or changed for a full paper:
    – Fig 1 and 2: Does the figure depict monthly volumes? How is the “search volume index” defined?
    – For reproducibility, how exactly was the web corpus of the French ana-mia community built? The sampling method (such as snowball) may strongly influence the structure of the constructed network and the selected nodes.
    – A quantitative support of the claim of “density..is globally low and locally high” could be given by the ratio of clustering coefficient to the clustering coefficient of a random graph null model (since the clustering coefficient alone is not very informative). Further, since the context of the paper is about information exchange within the community, it might make sense to measure the “information centrality”, which measures “how well the nodes of the graph exchange information” (see e.g. here).
    – The use of the diameter seems sub-optimal, since this measure is sensitive to chains (which can even be clearly seen in Fig. 3). Better would be either average path length, or for example an “effective diameter”, defined here as the interpolated 0.9-quantile of the diameter: paper (pdf)

    Ideas for possible further research which could make the paper stronger:
    – Community detection to see how e.g. modularity has changed over time, and why there are different communities (maybe corresponding to different types of disorders? Could health experts use this information for different targeting strategies?)
    – More detailed percolation study, how removal of targeted nodes affects the network structure, to find the optimal strategy of removing nodes.

  4. The contribution considers networks of ana-mia websites and examines the evolution of their topography when faced with actual or potential bans. The analysis is well conducted with an interesting case of networks dynamics. The presented findings offer valuable perspectives for the design of policy and communication strategies related to anorexia. The paper clearly belongs to a JITSO perspective as it combines both online and offline phenomena and focuses on some of their mutual interactions.
    I have just a couple of questions:
    The author highlights the fact that survivor sites are strongly connected with their online neighbourhood and refers to that as one of the “strategies” of the surviving blogs. One might wonder if the causal relationship could also be the opposite as the one proposed by the author (being more locally connected is a consequence of survival more or as much as a “strategy” that makes it possible). Are the data providing more info on this?
    The author states that the ban and the anticipation of the ban are “just as effective”. While this is a very interesting research question about how the web is also shaped by anticipations, are both cases really equal (with no differences in amplitude) or just similar (same pattern)?

  5. Pingback: Just-in-Time Sociology (JITSO) workshop just concluded « Paola Tubaro's Blog

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s