I've noticed increasing talk of civil war in the UK since the referendum:

The Lindy effect — assuming it's appropriate and that I've understood how to apply it for estimating how long things will last from now — says that if I assume I'm in a normal period of British history then the default risk of civil war in the UK in the next 5 years is ~(5 years)/(367 years) = 1.362%.

I assume that if people are actually talking about it, the risk is higher than normal. What I can't do is even begin to guess how much higher.

Is this just nut picking for the sake of exciting headlines, or is there a significant (let's say Pcivil war≥0.1) elevated risk? Is there any way to estimate what that risk might be? (Note that I can't go to the History exchange and ask "what are the general risk factors of all past civil wars" — I tried that, it was closed as too broad).

  • 45
    Before you answer: Note that the question is asking for methods to estimate the likeliness of a civil war. It is not asking about the current likeliness of a civil war in the United Kingdom. Please don't post answers where you try to make your own prediction in this regard, because questions about predicting the future are off-topic.
    – Philipp
    Commented Feb 12, 2018 at 10:02
  • 1
    I cannot find a clear question within this post. Can you please include/highlight the exact question you are asking?
    – Alexei
    Commented Feb 12, 2018 at 10:02
  • 2
    @Alexei I added emphasis to the question asked in the question body and edited the subject line to better reflect it.
    – Philipp
    Commented Feb 12, 2018 at 10:06
  • 4
    Usually civil wars are not recognized as such after a long time since they started; usually they just begin as protests/coup d'état/political terrorism that quickly escalate, in no little part due to the belief of the parts that they can win the conflict without it becoming a full fledged war. And, if it is difficult to recognize them when they are starting, it would be way more difficult to recognize them ahead of time.
    – SJuan76
    Commented Feb 12, 2018 at 10:26
  • 2
    @SJuan76 - yep, English Civil War(s) didn't exactly start out to be civil wars. Heck, most of them didn't want to get rid of the king for years (as in, even once the war started)
    – user4012
    Commented Feb 12, 2018 at 15:27

4 Answers 4


Predicting conflict, coups and civil war algorithmically has been done since the '90s. It was observed that human-based predictions were subject to various bias, (such as humans wanting to make their reports more exciting, attract funding and build prestige) So DARPA (a branch of the Pentagon) developed a fairly simple algorithm called the Integrated Conflict Early Warning System, that consistently outperformed human experts. It claimed over 80% accuracy source

Others have gone further. At Duke University, they use an ensemble model to forecast coups. It predicted the most likely places for coups to be Burundi and Thailand. It uses a combination of government characteristics (Stronger democracies tend to have fewer coups, but elections in weaker democracies can be a focus for coup attempts), economic characteristics (Hungry people riot), the past history of coups, and other social factors (for example infant mortality correlates to higher coup risk.) (source for list of covariates) Among the "Democratic Western Countries" the USA has among highest risk of coup in this model, due in part to the failure of the health system to reduce infant mortality compared to other Western Democracies. It considered the probability of a coup in the UK to be very low.

  • 1
    @JamesK I'm still not seeing the value in singling out the USA(or any country for that matter), when there are similar countries which rank higher (Belgium for example, if we go off that map). Especially when they admit that the increased risk in their model is a consequence of infant mortality, which has more causes than a 'failure of the health care system'. vitalrecord.tamhsc.edu/american-infant-mortality-rates-high Also, the actual list appears to be here github.com/andybega/mireg-blogs/blob/master/forecasts/… . Commented Feb 12, 2018 at 17:02
  • Comments are not for extended discussion; this conversation has been moved to chat. Commented Feb 13, 2018 at 23:59
  • 1
    Has anyone of these models been run for America in the last 6 months? Commented Sep 9, 2020 at 3:52
  • ote that the question is asking for methods to estimate the likeliness of a civil war. It is not asking about the current likeliness of a civil war in the <s>United Kingdom</s> USA.
    – James K
    Commented Sep 9, 2020 at 17:08

Tl;dr: The quotes are a mixture of provocation, sloppy word choice and metaphorical use. A true civil war is prima facie inconceivable in the UK. Missing root causes and missing recruiting potential make it impossible.

A note about forecasts. It is no accident that this is the term used for weather; there are strong similarities. We know that "prediction is very difficult, especially about the future"; while our factual knowledge as well as our theories are surely deficient, I suppose that the sheer complexity of the (political, sociological, technological, ...) system whose development is to be predicted severely limits the scope of any prediction in principle. Complex systems are often non-linear, or chaotic; small, early deviations quickly escalate and lead to completely false results for later points in time. E.g., a few thousand more votes in one state or another during a presidential election lead to an unexpected feminist anti-harrassment movement. Historical development has, underlying tendendencies notwithstanding, a strong chaotic element. Any attempt to attach a specific fraction of a percent likelihood to certain events is a fundamental conceptual error.1

The Term "Civil War"

Before we actually discuss possible causes for a civil war and investigate whether some of them are present in today's UK, let's first define what a civil war is (emphasis by me):

A civil war [...] is a war between organized groups within the same state or country. The aim of one side may be to take control of the country or a region, to achieve independence for a region or to change government policies. [...]

A civil war is a high-intensity conflict, often involving regular armed forces, that is sustained, organized and large-scale. Civil wars may result in large numbers of casualties and the consumption of significant resources.

To put this definition in perspective: According to Wikipedia, the Troubles in Northern Ireland are 'sometimes described as a "guerrilla war" or a "low-level war"', but do not appear to qualify as a civil war, probably because the high intensity was missing. After all, life mostly continued normally for large parts of the population. Contrast this with the situation in countries with present or past full-fledged civil wars, like South Sudan or Sierra Leone, which both have millions of displaced people and suffer from famines, disease outbreaks and other humanitarian emergencies.

For your question it is particularly important to distinguish the concept of civil war from other types of unrest like riots, terrorism, rebellions, coups etc. These lack organisation, scale, and/or are not sustained or intense enough to qualify as civil war.

With this picture in mind it appears that the talk of civil war in the journalistic texts you quote is a mixture of

  • provocation, especially when activists are quoted;
  • sensationalism, paired with sloppy journalistic craftsmanship — they didn't even read the Wikipedia article!;
  • metaphorical speech ("there is a civil war in the Labour Party").

Even before we explore causes of civil wars, a situation comparable to places with actual civil wars appears exceedingly unlikely for the UK.

Causes of Civil Wars

Common sense already dictates that a grave dissatisfaction is the root cause for any rebellion, let alone a civil war. The Wikipedia article lists a plethora of possible sources for dissatisfaction, material (hunger, poverty) and immaterial ones (religious and ethnic conflicts, dysfunctional administrations, bad governance).

Interestingly a third factor appears to play a large role: Civil wars are more likely when it is easy to recruit "foot soldiers". Contributing factors are poverty, a large, [young], badly educated population, and trivialities like rough terrain [probably resulting in weak governmental control].

It is obvious that very few if any of these causes apply to the UK, now and in the foreseeable future.

Yes, there are grievances and a sense of a lacking ability to participate politically; these are why we are having this discussion. But first of all the grievances are minor and, considering that the UK has free and secret elections, not comparable to the ones fueling civil wars in other parts of the world.

Secondly, all other causes are missing. While poverty is relative, the UK is one of the richest nations in the world. Even though large-scale changes in the political landscape happen fast compared to the second half of the 20th century — washing up inapt or dubious political personnel and generally contributing to a sense of insecurity —, the political institutions as such seem stable enough; administration and law enforcement are certainly working well. Education levels are high. The population is aging and shrinking. There is no way to recruit a rebellious army (of young men!) for any sustained amount of time.

While there is potential for religious tensions between Christians and Muslims, general consensus seems to be that a civilized way of living together is the British way to solve them. The terrain is not overly rough, with the exception of parts of Scotland (which indeed did have a civil war in he 17th century!).

As mentioned in the beginning, a civil war situation in the UK is already prima facie hard to imagine; after examining potential causes for civil war it is as good as impossible.

Rebellion, yes. Burning cars, terrorist attacks, riots: yes. But civil war: No.

1 It is already impossible (and, consequently, all attempts to do so have wrong results) to make an accurate prediction of the likelihood of a catastrophic failure for a single nuclear power plant. One such event in Fukushima significantly changed the course of Germany's energy policy. This is although all technical facts about the power plant are knwon and documented in great detail, and the theories of the nuclear and non-nuclear processes governing its behavior are complete and correct. (Contrast this with sociological processes.) Part of the equation surely are the human operators. The behavior of this comparatively well-understood, well-documented artifact plus human operators is already too complex to accurately predict under all circumstances.

  • Thank you for those good points. It made me realise something I had not noticed that I had forgotten: although I would be unsurprised by 100s of people in the UK finding guns and making a mess of things, there is an enormous gap between that and a civil war. Riots and arson (that's already been done this decade against Austerity) do not a civil war make.
    – BenRW
    Commented Feb 13, 2018 at 23:26

In order to determine the likelihood of a civil war breaking out or happening you have to first define what is a civil war?

If the government is militarizing peacekeeping forces to violently end protests and riots is that a civil war? What if the government is quietly but systematically labeling a significant portion of its dissenting population criminals, so they can strip their rights? What if that opposition claims the news cycle for weeks at a time? What if the government is seeing a large amount of homegrown "Terrorism" aimed specifically at government targets?

Some times its easy, the government has completely lost power, there is fighting among factions in the streets, and refugees attempt to flee. Something like has been happening in Syria for several years.

Other times its harder like what happened in Darfur. It was done quietly, and most of the world ignored the massacres. The signs were all there but no one bothered to look for them. And then there is a situation like the US Today. If we were seeing police forces roll in armed with combat gear and firing live ammunition at people protesting police violence on a pseudo regular basis; in other countries we would be talking about the sporadic civil war they were having.

So you determine the likelihood of it happening by defining the conditions that equal civil war, then based on events that have been happening, and how they can correlate between other events that have turned in to civil war, vs similar events that resolved peacefully or at least with out devolving into civil war.

Then the experts take those metrics and determine how likely a civil war is to happen based on historical context, the current tone of the opposition and how close that opposition is to actions that have proved, or are expected to prove a tipping point, and how likely based on current events those actions are liable to escalate to turn dissent into armed conflict. Then if that armed conflict should be considered a civil war, or labeled with some other less antagonistic term.

  • 1
    I agree that one has to define what a Civil war is (although I thought before that I have a fairly good idea of it involving guns and lots of shooting). However the answer falls short on the method for estimating the chances/risks of a Civil war. Some references to unknown experts but nothing more. Commented Feb 13, 2018 at 8:31
  • I think that a revolution and a civil war both have clear definitions for those who study history and political science professionally, unlike some of the hype loving media, who call every minor riot a revolution, or its suppression by a lawful government - a civil war.
    – TimSparrow
    Commented Feb 13, 2018 at 10:45
  • Does the Troubles count as a civil war by your metrics?
    – pjc50
    Commented Feb 13, 2018 at 17:01
  • @pjc50 - It depends how you define civil war. I am not dictating how anyone should define it. Commented Feb 13, 2018 at 22:18
  • @Trilarion - I provided a high level explanation of how it can be done. And it also explains why with out a greater detailing of the parameters providing a more specific method is impossible. Commented Feb 13, 2018 at 22:37

If you want a rough idea of the likelihood

A good first step might be to further filter the sources. You're already ignoring social media, since the signal to noise ratio makes it difficult to extract useful information. Apply that sort of thinking to the other sources as well.

Look at past predictions they've made, either explicitly, or more implicitly by publishing a disproportionate number of articles about the possibility that the event would happen (suggesting they believe it likely enough to be newsworthy, if nothing else).

Then look at the success rate, especially for events that seemed unlikely at the time. (and I cannot stress enough that it's the rate of successful predictions, not the number of successful predictions that matters here). If multiple sources with a past history of accuracy are making the same prediction, it might be worth paying attention to.

If you want an actual probability

Too bad.

If you just want an extremely rough estimate of the probability based on a decidedly oversimplified model

Good news!

You could then go a step farther and compute an actual probability using this method:

We first make the simplifying assumptions that each source has a consistent likelihood of being correct in its predictions, and that the sources are independent of each other (neither of which is entirely true, but we want to keep the math tractable here). We will also be ignoring any sources that didn't make a prediction one way or the other, as well as what happened when each source made the exact opposite prediction in the past. Including all these variables would improve the model greatly, but not by enough that I'm willing to take the time to deal with them.

For all sources i, let x_i be the frequency with which an event has come true when i made the same prediction it made this time. P(X | x_i) means the probability that X will happen given that x_i happens.

 P(X | x_1, ..., x_n) = P(x_1, ..., x_n | X) P(X) / P(x_1, ..., x_n)
                      = P(X) P(x_1 | X)/P(x_1) ... P(x_n | X)/P(x_n)

So to take an example, let's say we have three sources, and they're predicting whether Earth will be destroyed tomorrow.

  |                    | Pred| Freq | Happened
1 |       Daily Planet | Yes |   .4 |       .5
2 |           Quibbler |  No |   .9 |       .1
3 | New New York Post  | Yes |   .2 |       .6
  • Pred is the prediction they made.
  • Freq is the frequency with which they make that same prediction (i.e. how often do they say the answer to "will something happen" is yes/no?). This is P(x_i).
  • Happened is how often the event happened when they made that same prediction. This may be different for each source because they don't always make predictions about the same things. This is P(X|x_i).

And finally, we have P(X), which is the prior probability of the event happening. For this, we'll need to make an estimate using different techniques. So let's assume that all rocky planets have a similar likelihood of being destroyed on any given day (another blatantly false simplifying assumption, but just roll with it). The age of the solar system is about (365 * 5) billion days, and while we don't know exactly how many planets were destroyed, we're pretty sure the moon was born when two planets collided, and the asteroid belt came from somewhere, so let's call it an even 3 out of 1825 billion. But my uncle's coworker's friend's brother killed a goat this morning and said the entrails looked a bit portentous, and that's never led me wrong before, so we'll round that up to a 1% prior probability that Earth will be destroyed tomorrow. (For the civil war example, you might consider both the number of years in which the country was in a civil war divided by the number of years it's existed, and also the same figure for all countries, so you don't get stuck with a 0 likelihood for countries that haven't had any civil wars yet.)


P(X) = 0.01
P(x_1|X)/P(x_1) = .5/.4
P(x_2|X)/P(x_2) = .1/.9
P(x_2|X)/P(x_2) = .6/.4

P(X | x_1, x_2, x_3) = 0.0041667

So given those numbers, there's a bit over a 0.4% chance that Earth will be destroyed tomorrow.

Do note again that this will give a very rough estimate only. If you make different simplifying assumptions, or select a different set of sources to sample, you will almost certainly get different numbers. But it can offer a first approximation if you really want a specific number.

  • I think the problem is here that nobody knows the P(X | x_i) well enough. Hence the method is not applicable. Commented Feb 13, 2018 at 8:28
  • You forgot the step where you keep going back until your oracle box can output an exact fit to the past data that you used to make the oracle box. If it can match past predictions, it's future predictions will be extra predictive. Commented Feb 13, 2018 at 14:19
  • @Trilarion P(X | x_i) is the easy one. That's just "How often was source i right when it made a prediction in the past?" It's P(X) that's going to involve some guesswork. This is just a method to improve that estimate once you have it, based on the past reliability of the sources. But again, a rough estimate. My real advice is to stick to the first section where I suggested discarding sources unless they're right more often than they're wrong. The rest is just a way to get slightly better numbers when you insist on having numbers and the precise ones are very hard to come by.
    – Ray
    Commented Feb 13, 2018 at 15:20
  • @JackOfAllTrades234 Actually, you would want to avoid that in most cases; it leads to situations where the model can't account for the sort of normal variation that can be expected. Better to have a separate set of data that isn't used to train the model parameters, and thus can be used to estimate how well it'll perform in the future, and use that to determine when to stop training. But that would come into play for far more sophisticated models than this; here, we're just applying Naive Bayes since we don't have enough data to train a good model.
    – Ray
    Commented Feb 13, 2018 at 15:28
  • Is this based on an actual method someone is using to predict the occurrence of civil wars or is it merely a statistical procedure that could be used for such a purpose? Commented Feb 13, 2018 at 16:08

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .