« we-think and I-think | Main | book plug »

April 02, 2008



Ah yes, medical risk and mainstream journalism.

Obviously we should ignore the article. Why is that? Because:
1. Prof Wiseman's employer or other source of fame have no bearing on the truth of his statements. To think otherwise is the fallacy called 'argument from authority'.
2. Let us suppose there is in fact an excellent research paper indicating an increased risk. That does not preclude the possibility of another excellent research paper showing no increased risk at all, or even perhaps a benefit.
3. Let us suppose that a systematic review of existing research into the sausage/cancer link showed clear evidence of increased risk from sausage eating....

There are two kinds of risk. Relative and absolute. Suppose I have a 1% (absolute) risk of something. If I do something to halve that risk, I now have an absolute risk of 0.5%. I can express this in two ways:

ACME pills reduce your risk by 50% (relative reduction)
ACME pills reduce your risk by 0.5% (absolute reduction).

You can guess which figure looks better in print. But, when your absolute risk is already very small, even large relative reductions may be insignificant.

In a given year, your chance of getting bowel cancer is about 0.04%. If eating sausages increases that by a fifth (20%), then you can easily calculate you increase in absolute risk - a mighty 0.003% increase.

The 'meta-risk' is really just the multiplication of two risks, and is easily dealt with statistically.

Finally, statistical probability (is there another kind?) IS a form of knowledge. It's just not as detailed as some other forms.

If you pick a card randomly from a deck, I can say there's a 25% chance of it being a spade. That's real knowledge. It's not useless in decision making. Of course, if I'd memorised the order of the deck before hand, I might be able to say that there was a 100% chance of the card being a club. That's even more useful, but it's not some special new category of knowledge.

Risk is an odd thing. It cannot properly be used in decision making without some real understanding of it, but it's silly to write it off as just scientific mumbo jumbo that hides ignorance. Risk is present everywhere, because we don't have godlike knowledge of our surroundings. Most never think about risk in hard terms of numbers, but doing so can be very valuable, where those numbers exist.

Without a good understanding of risk and how to use it, you can't operate well in situations with a large amount of uncertainty but high upsides or downsides.

Those who wish to learn more about risk, should play cards with a pocket calculator nearby.

It is NOT the case that "knowledge would identify people who could eat dozens of sausages and not get bowel cancer". At best, further knowledge would simply allow us to produce more accurate risk figures, and it may allow us to identify groups where certain risk figures are close to 1 or 0 (within certain confidence intervals).

The only real foolishness is in clinging to fake certainty. All things* are risks. Its just that not all risks have been measured.

*baring purely logical constructs

Will Davies

J - I concede to you on virtually every count. To be honest, this isn't something I have thought about very much, and I was aware that there were some more considered views and ideas on the topic. So thanks for sharing some.

A meaker point, which I stand by, is as follows: the risk is a real thing, of which which we can have empirical knowledge (for all the reasons you lay out). However there are degrees of granularity. Estimating risk at the level of the individual will produce different results from estimating it at the level of the collective, hence arguments about collective v individualised health insurance. Treating risk as a collective entity is more efficient (for sort of Coasian reasons to do with transaction costs), but does not chime with the psychology of the individuals whose behaviour is being studied and altered. Therefore collective risks are falsely represented in terms that will be usable in individual decision-making.

I think something of the sort is going on with the sausages: they are aware that at the level of the collective, sausages increase the (relative) risk of bowel cancer by 20%, but do not or cannot drill down and work out which individuals are the ones responsible for this, so make a blanket warning.



Yes, you are right about granularity. When we say "British Males stand a 10% (or whatever) chance of dying of heart disease", we do NOT mean "Every who is British and Male will have a 10% chance of dying of heart disease". What we really mean is "10% of people who are British and Male will die of heart disease". The difference is, I readily grant, not at all obvious even to the quite thoughtful lay person.

The problem is that certain groups (not least the government) tend to leap from such facts about large populations to directives about what individuals should or shouldn't do. This is not helpful for the individuals.

If you make everyone excercise, it will not mean that everyone's risk of heart failure will drop.

The comments to this entry are closed.