[ExI] The Republican Party Isn't Really the Anti-Science Party

[rex]

Anders Sandberg <anders at aleph.se> [2013-11-12 22:46]:
>   Hurricane numbers are a bad indicator for climate: you get very few data
>   points even in a high hurricane year.
>
>   Consider flipping a biased coin, trying to estimate how biased it is. But
>   you only get 2-10 flips each year. Worse, you are trying to tell whether
>   the bias is changing. 
>
>   [ The variance of the estimate from N trials with pN heads is
>   (1-p)/(N+1)^2 (it is a beta distribution). going from 2 to 10 flips
>   reduces the variance by a factor of 0.67, which is just a 19% reduction of
>   standard deviation. If you want to reliably detect a change in p on the
>   order of 10% you will need a lot more data - at least more than 20 data
>   points. ]

Interesting idea, Anders, but I don't understand it. We have a record
of hurricane numbers h[i] for years i = 1:N, and temperature records for
those years. We could do a linear regression to try to estimate how
much of the variance in hurricane numbers is accounted for by temperature,
but where does the biased coin model enter into it?

-rex
-- 
Nostalgia is like a grammar lesson; you find the present tense and 
the past perfect.