Nassim Taleb on the black swans of war

[Admiral Yi]

No need to be a dick Guller.

[DGuller]

No need to be a dick Guller.

Too harsh?  Yes, in hindsight.  It was the "surely" that pushed the button.

[Admiral Yi]

You also did not consider the possibility that "are declining" can be interpreted in a number of different ways.

[DGuller]

You also did not consider the possibility that "are declining" can be interpreted in a number of different ways.

It shouldn't be interpreted in a number of different ways.  There is just one correct interpretation of the statement you bolded (regardless of the truth value of that statement).

[MadImmortalMan]

Kinda seems like predicting the outliers is more or less the point of insurance statistics. Sociologists and the like are more interested in broader trends, so ignoring them makes their work more applicable, at least in the short run. But the outliers usually have a bigger impact than the rest for good and ill.

You can't figure out who's going to win an election by counting outliers. But if you could predict an outlier like a tsunami, the impact of that one data point would be much much more impactful to the world.

[DGuller]

Kinda seems like predicting the outliers is more or less the point of insurance statistics. Sociologists and the like are more interested in broader trends, so ignoring them makes their work more applicable, at least in the short run. But the outliers usually have a bigger impact than the rest for good and ill.

You can't figure out who's going to win an election by counting outliers. But if you could predict an outlier like a tsunami, the impact of that one data point would be much much more impactful to the world.

No one predicts outliers.  Statistics is not fortune-telling.  I don't predict that John Doe will have a car accident on January 17th, 2016, or that a category 5 hurricane will make a landfall on Miami in 2018.  That's pure fiction, you need to be able to explain variance many orders of magnitude better than what is currently possible for that to be even remotely realistic. 

All statististicians just try to come up with probabilities, as precise as the data will allow them.  The difference between good and bad statisticians is that good ones don't overstate the precision that their data allows them.  Outliers are one of the variables that determine how much precision your data allows you, so ignoring them makes you overstate the precision, and thus makes you a bad statistician.

[jimmy olsen]

We're not trying to predict anything here, we're analyzing the past. Wars happened, or they didn't. People were killed or they weren't. Seems like simple arithmetic is appropriate for figuring out if wars and violence have declined over the last two thousand years.

[MadImmortalMan]

Kinda seems like predicting the outliers is more or less the point of insurance statistics. Sociologists and the like are more interested in broader trends, so ignoring them makes their work more applicable, at least in the short run. But the outliers usually have a bigger impact than the rest for good and ill.

You can't figure out who's going to win an election by counting outliers. But if you could predict an outlier like a tsunami, the impact of that one data point would be much much more impactful to the world.

No one predicts outliers.  Statistics is not fortune-telling.  I don't predict that John Doe will have a car accident on January 17th, 2016, or that a category 5 hurricane will make a landfall on Miami in 2018.  That's pure fiction, you need to be able to explain variance many orders of magnitude better than what is currently possible for that to be even remotely realistic. 

All statististicians just try to come up with probabilities, as precise as the data will allow them.  The difference between good and bad statisticians is that good ones don't overstate the precision that their data allows them.  Outliers are one of the variables that determine how much precision your data allows you, so ignoring them makes you overstate the precision, and thus makes you a bad statistician.

I imagine in your business, the outliers are things like car crashes, house fires, crimes and stuff. Don't you guys at least try to budget for the claims?

[DGuller]

We're not trying to predict anything here, we're analyzing the past. Wars happened, or they didn't. People were killed or they weren't. Seems like simple arithmetic is appropriate for figuring out if wars and violence have declined over the last two thousand years.

No, we're trying to figure whether the world in the last 7 decades was an inherently more peaceful place, or whether it was just lucky.  And we are in a way trying to predict something here, because by trying to suss out a trend, we're also trying to project what the environment will be like in the future.

[DGuller]

I imagine in your business, the outliers are things like car crashes, house fires, crimes and stuff. Don't you guys at least try to budget for the claims?

All those things aren't really outliers.  On an individual level, those are very rare events, but on aggregate the number of car crashes or house fires is quite predictable, since these events are uncorrelated, and you have a lot of policies.  Budgeting for them is not that tricky.

The outliers, or rather black swan events, are large scale catastrophes like hurricanes or earthquakes, or worse yet, another kind of catastrophe that you can't even think of yet, but whose losses you still have to cover.  There are catastrophe models out there, but the nature of fat-tail events is such that you can't ever truly know whether those are good or junk.  How exactly do you test whether the model accurately predicts 1-in-1000 year hurricane events?  All insurance companies accept a small risk that they'll just be shit out of luck sometimes.

[Zanza]

I find the general premise that wars are random events questionable. The long peace theory is not based on mathematical or chronological considerations, but on alleged or perceived changes to human interaction in the last decades. As human interaction is not random, but based on will a pure mathematical analysis could fall short.

[Eddie Teach]

Guller lacks the confidence in his models to set up a small self-sustaining scientist colony in Greenland to weather the coming storm and lead our descendants back to civilization.

[grumbler]

I find the general premise that wars are random events questionable. The long peace theory is not based on mathematical or chronological considerations, but on alleged or perceived changes to human interaction in the last decades. As human interaction is not random, but based on will a pure mathematical analysis could fall short.

Agreed. That's the problem with anyone looking at things from the standpoint of their profession; they assume they can answer the question using their professional tools.  As the saying goes, "to the man with the hammer, all problems look like nails."  Taleb's entire argument is premised on the assumption that wars are random events subject to statistical analysis.  To paraphrase him again, "just laugh at the person."

[frunk]

I find the general premise that wars are random events questionable. The long peace theory is not based on mathematical or chronological considerations, but on alleged or perceived changes to human interaction in the last decades. As human interaction is not random, but based on will a pure mathematical analysis could fall short.

Actually there has been at least one statistical study that claimed death by violence has been on a long term decline, and I believe that there has been a similar one for war.  You may not be using mathematical models, but other people have been.

[DGuller]

I find the general premise that wars are random events questionable. The long peace theory is not based on mathematical or chronological considerations, but on alleged or perceived changes to human interaction in the last decades. As human interaction is not random, but based on will a pure mathematical analysis could fall short.

Agreed. That's the problem with anyone looking at things from the standpoint of their profession; they assume they can answer the question using their professional tools.  As the saying goes, "to the man with the hammer, all problems look like nails."  Taleb's entire argument is premised on the assumption that wars are random events subject to statistical analysis.  To paraphrase him again, "just laugh at the person."

I think this criticism misses the mark as well.  Statistics can be and is applied successfully to many events that are not truly random (if there even is such a thing as a truly random event).  Car accidents are not truly random either, they're a result of two drivers' decisions that eventually made them occupy the same space at the same time.  The key is whether the chain of events is so complex that the outcome is not truly predictable due to butterfly effects except in extremely short time horizons.