Heroes Not Zombies

Just read Reckoning with Risk by Gerd Gigerenzer (ISBN 978-0140297867) and it’s one of those books that has changed the way I think.

It’s about the way statistics are presented to us to create an illusion of certainty. We don’t live in a predictable world. In fact, Gigerenzer uses a quote from Benjamin Franklin (there he is again!) about certainty –

In this world nothing can be said to be certain, except death and taxes.

He refers back to this from time to time throughout the book to remind the reader that when something is presented as certain, it probably isn’t.

Why would statistics be used to present us with an illusion of certainty? Well, two reasons mainly. Firstly, we can’t handle too much uncertainty. We need some predictability in our lives. If we really felt absolutely nothing was certain we’d be paralyzed. Every morning when I get on the Glasgow train I’m pretty sure I’ll end up in Glasgow (of course, being Scotrail, I’m not at all certain exactly when I’ll get to Glasgow! But I’m pretty sure I’ll get there all the same). Even though I have this degree of daily certainty I do know, somewhere in the back of my mind, that accidents happen, that people get sick and that one day I’ll die. And I have no way of knowing if any of those things will happen to me on the Glasgow train today. But if I tried to base my daily decisions on all of those possibilities I guess I wouldn’t even be getting on the Glasgow train!) The second reason, is that others – experts, organisations, authorities and companies – want to exert their power over us. (see how to make a zombie).

Gigerenzer’s response to this is education. His book illustrates how we are all innumerate. He bases the whole book on a small handful of scenarios which makes the book both easy to understand and quick to read.

Here are the two main things I learned from him.

We understand frequencies much, much more easily than we understand probabilities. Try this out on your friends (especially doctor friends) – here are two ways to present the same information about mammography –

The probability that a 40 – 50 year old asymptomatic woman has breast cancer is 0.8%. If she has breast cancer, the probability of a positive mammogram is 90%. If she doesn’t have breast cancer, the probability of a positive mammogram is 7%. Imagine a woman with a positive mammogram. What is the probability she has cancer?

When Gigerenzer tried this out on experts very few got it right! He then showed them this version –

8 out of every 1000 women has breast cancer. Of these 8 women with breast cancer, 7 will have a positive mammogram. Of the remaining 992 women who don’t have breast cancer, 70 will have a positive mammogram. Imagine a sample of women who have positive mammograms. How many actually have breast cancer?

See how easy the second example is? This is a very good mental tool for clearing away the confusion created by probabilistic statistics. He shows how to make a decision tree using this method. This is going to make it much easier to understand clinical trial results for me. How come I wasn’t taught something about this at Medical School?

The second lesson is more important because its about how to see through attempts to manipulate us with statistics. This is slightly more technical – there are three ways to present a comparison of two groups of people who have had, say, either two different treatments, or one group gets a drug and the other placebo. The three ways are Absolute Risk, Relative Risk and Number Needed to Treat. Let me quote one of his examples.

The West of Scotland Coronary Prevention Study published a Press Release about the use of statins (the lipid lowering drugs). It said that from a study comparing a particular statin to placebo is was shown that taking the statin “reduced the risk of death from coronary disease by 22%” – well, that seems pretty convincing doesn’t it? But look at the actual study. It compared two groups of 1000. One group got the statin. 32 of them died. The other got placebo and 41 of them died. The absolute risk is the proportion who died in the placebo group minus the proportion who died on the statin. That’s 0.9%. The relative risk is the absolute risk divided by the proportion who die in the placebo group. That’s where the 22% figure comes from. The Number Needed to Treat is 111. That is, that you need to get 111 people to take the drug for one of them to get the benefit of not dying. Well, I’m sure you’ll agree, not all three of these presentations seems the same. Gigerenzer shows how drugs companies and authorities routinely use Relative Risk to emphasise the potential benefits of their treatment while at the same time presenting the potential harms as Absolute Risks to minimize the impression of adverse potential. This is just manipulation. He makes a very good case for why we all deserve to give informed consent to treatments but how we rarely get the chance to be properly informed.

He makes the interesting point that experts often claim that their tests are absolutely certain – whether its cancer tests, HIV tests or DNA matching, but shows us how that cannot be so. There really are no such certainties and we shouldn’t believe anyone who claims otherwise.

I’d recommend this book. You won’t hear TV news the same way again. He’s right. Education is our way of liberating ourselves from the agendas of the experts and so called authorities. We should make up our own minds and learn how to do that.