On Certainty

If you're a scientist or know anything about science you will probably want to stop reading this right now. I'm about to try to explain alpha errors and will almost certainly make a has of it.

Anyone who has been reading my latest entries will know I'm going a little crazy trying to figure out how much attention I should be paying to the scientific doomsdayers who say a + 4º planet is not fit for human inhabitation and would therefore result in Near Term Human Extinction. I was trying to talk to someone about it and, my first advice, don't do that. No one wants to hear about it. Literally no one. People have regular lives with a million little things built into them specifically so people can worry about sensible things and not craziness like human extinction. People worry about their kids, the schools their kids go to, the economy, their pensions, their next week at work, the state of their mortgages and other debts, many many normal things. They really don't need a lunatic like me dumping human extinction onto the pile and upsetting everything. Second thing, the concept of certainty requires explanation.

This is my best guess because I am not a scientist. I took some science as an undergrad but not much. Just enough to understand alpha and beta errors. So when scientists say they are 95% certain something is happening, it means they know something is happening and they are 95% certain it means what they say it means. Or, if they are predicting, it means they are 95% certain about their model. There is a huge difference between measuring and predicting (obviously) but the more important question is, "How certain is 95% certain?"

That 95% is a very important number because by itself it tells us a lot about the state of climate science. It is the statistical representation of the probability of a false positive. That's what alpha error is, a false positive. Reporting something is happening when it isn't really. The kind of opposite of alpha error is beta error and that's a false negative - reporting nothing is happening when something really is. No one needs to worry about beta error at all so long as we are talking about climate change because everyone agrees something is happening.

Alpha is typically expressed as a decimal but it gets converted into a percentage for public consumption. When I was an undergrad, we were taught that in most cases an alpha of .2 was acceptable. Meaning the results are accurate 4 times out of 5 and the corresponding certainty was therefore 80%.  If you wanted a higher degree of certainty, you could make alpha .1 (or 9 times in ten, 90% certainty). So a 95% certainty is an accurate result 19 times in 20 or an alpha of .05

When you express it as a percentage it completely changes how people will perceive the number. 95% certain seems 5% more certain than 90%. And it is, kind of. But 1 error is 5 is twice as many errors (statistically) as 1 error in 10. And 1 in 10 is twice as many as 1 in 20. So 95% certain is not 15% more certain than 80% certain -it's four times as certain.  An alpha of .2 is four times as high as an alpha of .05 - and that's the alpha they are using.

So when climate scientists say they are 95% certain something is happening, they mean they absolutely and conclusively know something is happening but there is a 5% chance their results are occurring because of a selection bias.

I don't know how climate scientists deal with selection bias. When I was dealing with alpha and beta errors the biggest problems were geographical and historical - people in different parts of the world are different and that causes problems for generalizing results in the social sciences. But weather is always different in different places. It has to be. That's the definition of weather and it happens because we have an atmosphere (without which we would be even more completely fucked than I think we are).

I'm going to try to get an expert to explain this to me and I'll report back.