Risk and Probability

Nassim Taleb's take on risk and probability. 

Discuss in comments how this applies to statistical approaches to terrorism and insurgency.

Comments

[russ greene]

Risk management tends to focus on quantification (using past frequencies of various types of events to predict future frequencies).

The problem is that the events that are most likely to be catastrophic, are also likely to surprise us for qualitative rather than quantitative reasons. That is, they are not likely to fit into any category that we have been keeping statistic records of.

[Alex @ I-Con]

This does not apply directly as he seems to be talking about normal curves which only work for continuous variables and seemingly random events such as stock fluctuations over the course of the day.

However, the idea may work. When an event is very rare, such as a terrorist attack on the US, we cannot properly predict it. Such a small data set creates massive uncertainty. This is one of the major challenges in designing analytics.

Let's take the case of algorithms designed to identify terrorists based on transactions, e-mails, etc. According to Baye's Theorem of conditional probability, as the probability any subject being watched is a terrorist is very small, even if the chance of a false positive is one percent or less (it won't be if we want to be relatively sure we identify the terrorists) there will likely be millions of false positives for every uncovered plot. I plan to write more on such challenges in the future.