Tuesday, 20 April 2010

Risk versus Uncertainty - in the skies over Europe

Planes are grounded yet again across Europe as volcanic ash and particulates disperse into the atmosphere out of Iceland. The "risk" to passenger safety is too great, so the authorities issue grounding orders across most of European airspace.

By all accounts the risks associated with flying an airplane into a cloud of volcanic emissions (catastrophic engine failure) are not worth the potential benefit (getting from A to B quickly). Hence, preventing planes from flying into such clouds is an extremely sensible risk mitigation policy. No argument. Then why are we starting to hear grumblings from airlines, complaining that they are being grounded unnecessarily?

Is it because they are just trying to turn a buck and there isn't much demand for a ticket on a plane that can't take off? Definitely.

Is it because airlines, driven by their profit motive, are more cavalier when it comes to risk? Almost certainly not.

So what gives?

It's all about risk versus uncertainty. So it is worth going back to basics to understand why these two animals are related, but different.
  • Risk is the extent to which any particular event may have multiple outcomes.
  • Uncertainty is the extent to which we have incomplete knowledge about the nature, range or relatively likelihood of these multiple outcomes.
For some people that is a "so what" proposition. However, I am pretty certain that for the overwhelming proportion of the population that is a "huh?" proposition. Assuming the former have already moved on, I will expand further for anyone interested in reading more.

Let's proceed by contrasting two examples. Let's first take a coin toss. Say we bet on a coin toss, so our fortunes are inextricably tied to the outcome. The toss exhibits "risk" for us, there are multiple potential outcomes - heads we win, tails we lose. However, there is relatively little "uncertainty" - a decent spinning toss should have a 50% chance of coming up heads and a 50% chance of coming up tails (rounded to exclude the odds of it landing on its edge). We can be pretty certain of those probabilities, if we know who is tossing the coin, how the coin is being tossed, the nature of the coin being tossed - these are all regular, observable (knowable) qualities. This example exhibits a lot of risk for us in our bet (one chance in two that we lose), but relatively little, if any, uncertainty (we are pretty confident that it is at least very close to a 50/50 chance)

Now lets look at flying in a plane. To simplify the infinite number of potential nuanced outcomes; the plane may complete its journey or the plane may crash. So, like the coin example, our next step is to try and quantify the relative likelihood of the potential outcomes (assuming we have correctly identified all relevant potential outcomes). In this situation the airlines and regulators have made decisions and taken actions that are designed to affect the relative likelihood of the outcomes (crash/not crash in our example). The plane has engines for example - always a good start. The pilots are trained professionals - always a bonus I find. There are copious check lists and scheduled maintenance routines. And so on and so forth. Moreover, we have emprical data for millions of commercial flights and only a very small proportion of those have resulted in crashes. So we are left with an estimate of the probabilities of Very Low (crash) versus Very High (not crash).

Where are the numbers, the hard probabilities this time I hear you ask. Well, I can't give them to you. Why? Because I am uncertain of all the various things that could possibly go wrong, be completely unanticipated, or are just so absurd as not to waste time considering them (shot down by an alien spacecraft). Such is uncertainty. In technical-speak, I am unable to specify the probability distribution function. In this example, we have low risk (the odds of crashing are estimated to be extremely low), but we have a lot of uncertainty (is the probability of a crash 0.01% or one hundred times greater at 1% for the specific flight in question? - we can't tell. It may in fact be 50% if someone happened to leave the work experience kid to do the pre flight engine check).

So what are the implications of this and how does this relate to plumes of volcanic ash?

Well, first off, note what is the most important implication of this understanding for risk and uncertainty. We act to mitigate against risk - that is try and prevent the known potential bad outcomes occurring. The more successful we are at doing that, removing risk, the more we are left with uncertainty. People who are scared to get on a plane because they are scared of the "risk", are actually scared of the "uncertainty" - the unknown or unknowable factors that might be at work. And you can't do anything about uncertainties directly - obviously because you don't know about them. And there is the crux:

Uncertainty is just a condition dependent on how much information we have - remember that because I will return to it.

So, on to the ash and trying to understand why airlines seem willing to fly their passengers into ash plumes, while regulators aren't. It isn't profit motive per se, although it must be accepted that the regulator isn't affect by whether the planes fly or not. In this particular instance there is specific cause to believe that a tussle between risk and uncertainty is taking place and one party might be confusing the two concepts.

According to reports, the Europe wide ban on flights is largely the result of the work of the UK Met Office. They model the atmosphere and predict where the volcanic emissions might drift. They come up with a result and recommend that flights be banned in the areas around which their model suggests the ash will be. Very sensible in itself, it is an attempt to assess the risk. However, what this work does not incorporate is physical measurements of where the ash actually is at any time. So, there is also significant uncertainty at work here.

Planes are being grounded because of the risk (the estimated odds), certainly. But, they are also being grounded because of the uncertainty (the not being confident of the estimated odds). Now, this is where things could possibly end. When we make such a decision, the uncertainty needs to be considered in conjunction with the risk and combined they suggest that it is prudent to ground planes.

So the airlines are wrong you might say. Well, not necessarily. What the airlines are validly pointing out is that the uncertainty part reflects our lack of sufficient information, which affects our confidence in assessing the risk. And they go on to point out that there is more information that could be obtained. That information is where is the ash in reality (as opposed to where it is modelled to be), which they suggest can be obtained by making measurements, possibly by flying suitable test flights in certain areas, or using balloon based sensors maybe. The simple analogy is that you don't put on a heavy coat when you leave home simply because you are uncertain about whether it is cold or not - you take whatever practical steps you can to gather a bit more information, like walk outside, so you can remove more of the uncertainty and leave yourself with the risk that it may or may not get cold later.

The airlines are saying that some of the uncertainty, which is part of the reason for the blanket flight ban, is not because we lack sufficient information, but because we aren't even bothering to gather the relevant information. And that is a valid point.

So who is right? Well, both are and neither are. We just don't know. We are dealing with uncertainty, remember? But one thing is certain. If we understand the difference between risk and uncertainty, we can make better decisions. In this case minimise the possibility of a crash, but also minimise the possibility that we are unnecessarily curtailing air transport.

1 comment:

Anonymous said...

I think that the difficulties with applying the theory of probability to the atmosphere is actually a difficulty with handling its advanced complexity - the reactions are very very many in number, and the character of these reactions is each probabilistic rather than deterministic. (Rather than a difficulty in measurement.)

Surely, there is an analogy between the two systems.

However, I think that the reason that there is less uncertainty in economics and finance is as a result of the twin approach of (a) the ability (and effort) to increase the variety (complexity) of the regulatory forces, and (b) decreasing the variety (complexity) of economic life through legislatively reducing it to match the regulatory environment.

You can't do this with the atmosphere!

There is another point to be made about how the drive to reduce uncertainty in economic life will tend to reduce the variety (I'm using the term here as a measure of complexity) of economic life - and it is from this variety that we get true and powerful innovation, craftsmanship, different forms of organisation and what have you.

These are the kinds of things that genuine economic recovery has grown out of in the history of these crises - I fear that the increased regulation of economic life we are currently experiencing may have the effect of subverting these things.

It's a difficult balance, for sure.