[Question] Where to find Base Rates?

[adam demirel]

As we are all aware from Thinking, Fast and Slow, one of the largest decision-making errors we make is the Base Rate Fallacy. Just to remind you—the error is that we don’t correctly apply probability reference classes to ourselves.

Two examples:

• The likelihood of winning the lottery is x, but I am more likely to win than x.

• 1 in x people will get lung cancer from smoking, but I won’t.

This is all well and good, but my question is where you go about actually finding x. In other words—where do you find your base rates? Any particular repositories of knowledge that we should know about? Any processes that we should follow?

Let’s say you’ve decided you want to reduce your risk of getting Coronary Heart Disease. Is your first stop to go to WHO and gather their list of recommendations, or something else?

[Bucky]

I often use the Office for National Statistics (UK)

[adam demirel]

Interesting. So let’s say you wanted to minimise your risks of CHD as the OP mentioned—you would just go here and find out the highest correlates and trust them enough to base your whole gamble on?

[Bucky]

To be honest I’d just google that one but that didn’t seem like very useful advice! My googling got me almost straight to this risk calculator used by NHS Scotland. Cross check this with a few other references from google and that’s probably as good as anything I’d work out myself by going to the data—it’s a well studied issue.

ONS is useful for base rates where google fails me.

[habryka]

Sadly not answering the question, but I have the domain baserates.org, which I bought a while ago in a distant hope I could build a website that just has a large database of baserates for all kinds of stuff.

If anyone wants to take on this project, I am happy to provide the URL.

[Pattern]

I think this could work as an “answer” instead of a “comment”, since it’s part of a future solution (as opposed to a pre-existing one).

[Dagon]

I don’t have any recommendations on this, but I will warn you that the hard part is NOT finding out the numeric base rate for an easily-identified and well-studied group. The hard part is determine what is the appropriate reference class to use, and then mapping that into the available rates for slight or major variations from your preferred reference class.

When you want to know “how much will smoking this cigar increase my risk of cancer”, it’s very hard to find studies or estimates of people sufficiently similar to you making decisions on that small a scale. You can do some endpointing (find the highest and lowest plausible effect sizes from available groupings, even if you’re not a perfect fit for any of them), and for many choices, this is good enough—even very low estimates of effect may be strong enough to make your choice obvious.

Also, don’t completely discount analysis and inside-view. Your chance of a single ticket winning the lottery is directly calculable, regardless of how other people have fared in other lotteries.

[adam demirel]

I totally agree with you that the reference class problem is a real problem, but having access to a process for quickly finding accurate base rates is still a problem.