Why Switzerland is Peaceful

[This post is a reboot of a post originally published on my personal blog on 31st April 2014]

One of the hardest parts of academic research is the timeline for a project to go from conception, through initial exploration, rethinking, compiling results, writing up, submission to a journal, review, revision and finally publication. What invariably happens is that the work, that was for so long all you thought about on a daily basis, fades from memory to the extent that when it finally appears in print, it’s hard to remember what you actually did! For this reason I decided to write an informal blog post to remind myself and others about my work that recently appeared in PLoS One, a full 4 years after I started work on it and nearly 5 years since it was first started. This is a unique paper that I want to be as accessible as possible.

Map of religious and linguistic groups in Switzerland. Reproduced from Good Fences: The Importance of Setting Boundaries for Peaceful Coexistence (2014) PLoS One

The Short Story

The short story is that this paper uses a quantitative model for the likelihood of ethnic violence, which made accurate predictions of where violence occurred in the Former Yugoslavia, based only on data on where people of different ethnicities live. In order to make sure that this model is robust and universal, it should also be able to explain why somewhere like Switzerland does not have ethnic violence despite religious and linguistic diversity. When the effects of boundaries between groups are included in the model, whether they are the canton boundaries that allow for a strong federalism, or the steep mountains of the Alps, we found that these are enough to mitigate the tendency to inter-group violence that would otherwise be present.

What the Model Got Right (& What No Model Can Ever Get Right)

In the paper we found that Switzerland contains religious and linguistic diversity in such a configuration that the population has a high propensity to violence. However, consideration of geographical boundaries between linguistic groups and administrative boundaries between religious groups are sufficient to reduce this propensity to violence below a level that was found to lead to violence in the context of Yugoslavia. But what is remarkable is that we found several regions of the country where the violence remained at a relatively high level, specifically in the particularly pluralistic canton of Graubünden and the canton of Bern. Looking more closely at Graubünden, we see that Graubünden introduced unique sub-cantonal divisions known as Circles or Kreuze, purposefully to overcome religious tension. Considering these additional divisions reduced the propensity to violence to peaceful levels as observed. In Bern, the points of high propensity to violence corresponded very closely to reports of arson and even bombing, although not full conflict. Returning to Yugoslavia, we found that the presence of an administrative boundary between Slovenia/Croatia and Macedonia/Kosovo actually reduced violence since it coincided with a boundary between ethnic groups. It was elsewhere that violence occurred since boundaries were ineffective there.

Our model only describes an underlying tendency to violence which may not necessarily equate to realised violent acts within a given period of observation. Some trigger is inevitably required to incite actual violence, this is included as a variable threshold in our model. It is beyond the scope of this work, and perhaps any work, to predict exactly when a seemingly trivial event might spiral into severe conflict (the consideration of the final state of a system with respect to small changes in its initial conditions goes into the fascinating realm of Chaos Theory). Additionally, it is reasonable to expect that the threshold to violence be higher in a country such as Switzerland which is generally prosperous with a strong social contract.

Why These Findings Matter

The causes of ethnic conflict are massively complex and are in many cases entangled with very deep rooted historical, cultural, political and/or economic considerations that are specific to each country context. No single qualitative or quantitative model could ever explain all of these considerations completely. However, what is remarkable is that our model demonstrates that a single parameter characterising the typical spatial size of homogeneous communities can go very far in explaining the underlying propensity to violence between groups. This demonstrates a very deep analogy between social systems and physical systems such as molecules of condensed matter. Both have large numbers of components but the system as a whole may be described statistically by a single order parameter. The parallels between physical and social systems are fascinating; the curious blog-reader should delve into Phillip Ball (or Ilya Prigogine and Duncan Watts among others more technical treatments).

From a humanitarian and political perspective these findings are also extremely important, if a little hard to accept from a liberal point of view. One implication is that peace can be maintained if different communities are kept in isolation from one another. However this is only one implication of several findings. The exact shape of the groups is significant; the same population may be configured slightly differently leading to a lower propensity to violence than another. Therefore it is too simplistic to conclude that partitioning people based on ethnic identity is the only mechanism to eliminate violence.

We also demonstrate that peace can be maintained when groups are granted self-determination and governmental independence. Finally, we see that peace can be maintained if communities are well mixed; the multi-cultural model that has been adopted by Singapore since 1986. The case study of Yugoslavia demonstrated that violence occurred there only when national borders did not align with the boundaries between ethnic groups.

The role of identity was addressed incrementally and over a long period of time in Switzerland, leading to canton boundaries which were carefully chosen and incrementally adjusted. However, the post-Soviet, post-Ottoman and post-colonial eras led to many states formed with scant regard for the identity of its sudden citizens in Africa, Central Asia and the Middle East. In some cases the borders were chosen arbitrarily, and in others maliciously to prevent supremacy of particular groups. To further the analogy with statistical physics, these citizens were hastily quenched into arbitrary borders in contrast to the gradual annealing process of canton formation in Switzerland. The current conflict in Syria, for example, can be seen clearly as the breakdown of the hastily partitioned Levant, as decreed in the infamous Sykes-Picot agreement of 1916. The partition between French and British mandates allegedly determined by British officer Sykes as a line from the ‘e’ in Acre to the last ‘k’ in Kirkuk on his map.

The Intuition

The model begins with a hypothesis that the propensity to conflict between group A and group B is dominated by territorial considerations. If group A and group B are very well mixed then neither group can reasonably claim ownership of that space. Conversely, if the groups are very well separated then neither has their territory imposed upon by the other. This implies that there is an intermediate group size under which groups are large enough to gain a local majority and close enough to engage in violent acts. This can be formalised mathematically by using what is known as a wavelet filter, a mathematical technique used in image processing to detect objects of a given shape and size.

Ricker wavelet filter for detecting objects in images

Imagine applying this filter to a black and white image where black pixels represent ‘1’ and white pixels are ‘-1’. If the image is completely white, the positive and negative parts of the filter will cancel out. But if a black blob is placed in the otherwise white image, when the filter is placed on top of the blob a large signal will be returned if the blob size is close to 2r_{c}. This filter represents a group of one type of size r_{c} surrounded by a group of a different type. By comparing the actual distribution of groups in space to this template, we can measure the propensity to violence at that point. If the typical group size is much larger or smaller than the size of the filter, r_{c}, then the propensity to violence is close to zero. This is the basic principle behind pattern recognition; apply a filter of a given shape to an image and a strong signal will be returned when the image contains a feature that is the same shape as the filter.

In this picture, the definition of ‘intermediate sized’ corresponds to our chosen value of r_{c}. When the size of our different groups approaches this, then they are in a configuration prone to violence. We varied this parameter to investigate how it changed our results, but the findings were quite robust. In fact the best agreement was with a value of r_{c}=24 km which corresponds roughly with the distance that can be covered on foot in a day. That our model performs best with an intuitively sensible value for this parameter further demonstrates its robustness.

Future Work

This work has demonstrated that the dominant mechanism of mediation of ethnic conflict is in a spatially local fashion. In other words people’s inclination to violence is determined mainly by the people that they can interact with physically. One possible extension is to see how this changes in a context where internet mediated communication effectively changes the role of distance (a very interesting body of work has demonstrated this, see here and here). By using a spatially embedded network, it could be possible to see if the numerous, although weak ties mediated by social media and social networking lead to more or less volatility.

Finally, the application of this work is hampered by the fact that for many parts of the world which are of interest, reliable census data to be fed into the model is not available (for example, no full census has been conducted in Lebanon since 1932). In fact, addressing the data gap is something that I do in my current role for United Nations Global Pulse albeit for developmental purposes and not for analysing conflict. An interesting extension of this work could consider the GREG dataset, a painstaking digitisation of a series of maps containing 929 distinct ethnic groups compiled by cartographers at the Institute of Ethnography in the USSR Academy of Sciences. Since this remains arguably the most comprehensive census of some parts of the world, ethnic faultlines could be identified that persist today.

Data, science, data science and trace amounts of the Middle East and the UN

Love podcasts or audiobooks? Learn on the go with our new app.