Transcript - Briggs, 2015 - The Crisis Of Evidence, Or, Why Probability & Statistics Cannot Discover Cause - YouTube

Speaker1: [00:00:00] I'm going to talk to you about a revolution first and philosophy. My talk is basically half philosophy, half statistics. The title of the talk is The Crisis in Evidence. And what I basically want to do is I want to show you that probability and statistics have overpromised and have led to basically a pandemic of over certainty, particularly in what we might call the softer sciences, and in particular those fields that use these types of things to regulate us.

<aside> 💡 So basically what I want to tell you is that probability and statistics cannot do what they promise to do, in its classical sense, and that's to show causation.

</aside>

<aside> 💡 And that's the philosophical topic. And I want to explain that first, and then I'm going to show you that even if we assume that probability and statistics can show causation, even if we do understand causation, the procedures that we use are wrong and they should be adjusted and done in a completely different way.

Cause | causation | causality | BSI

And that way is essentially just what Ed was telling us. We replicate, we replicate. We have a model, we make predictions. We see if those predictions are upheld, and we have to do that repeatedly.

The problem with probability and statistics is they seem to show us, give a shortcut. They seem to promise that we could know things with very little effort. And I'm going to prove that to you.

</aside>

<aside> 💡 So what are traditionally probability and statistics used for? What do you think they're used for?

**Explain or quantify uncertainty in that which we do not know. And nothing else.

What is probability? | BSI**

</aside>

<aside> 💡 Strangely, however, classical procedure in both its frequentist and Bayesian procedures, say the opposite.

</aside>

So let me give you a little example.

We'll work with the PM 2.5 example. This is a I'll use a fictional example at first which is based on reality and then I'll use one that was actually used by the California Air Resources Board that I originally got interested in from Jim Enstrom here.

So imagine we have two groups of people who are exposed or not to PM2.5. A thousand people had either no exposure or just trace exposure and five of them develop cancer of the albondigas. And in another group, another thousand people which are exposed to some high level, Every single person in this high group: Each 1000 of each 1000 persons was exposed at the same level, and 15 of them develop cancer of the albondigas. So five and 15. Now, let me ask you this question. I don't know how many statisticians are on the audience, but what is the probability?

in a low or no group of 1,000 5 people got cancer of the albondigas, and in the “some” or high PM2.5 group of 1,000 15 did