[twic]

Something i find really fun is that every specific field of study has developed its own completely general statistical tools. There's no reason they couldn't be used in other fields. They just aren't.

Geography apparently has Kriging:

https://en.wikipedia.org/wiki/Kriging

Economists have LOESS (okay, used beyond economics, i admit it):

https://en.wikipedia.org/wiki/Local_regression

Maybe geographers refuse to use LOESS because to them, that's a boring rock?

Astronomers have sophisticated deconvolution algorithms that are completely unrelated to the ones microscopists use, etc.

[astro123]

Kringing is the same thing as gaussian process regression, and astronomers use GPs a fair bit. I'm not sure whether they are used more widely.

My favourite forgotten/isolated statistical method is MCMC. These were first used by nuclear physicists at Los Alamos in the 40s/50s, but weren't really recognized more widely until the 80s. This is probably partly because only people working on bombs had access to the computing power before then, but still.

[jwuphysics]

Astronomy is also an interesting case because this law doesn't necessarily hold true. The power spectrum, or Fourier transform of the autocorrelation function, isn't monotonic for matter on large scales. For example, baryonic acoustic oscillations at early times in the Universe get "imprinted" into the cosmic structure, such that galaxies have a preferential separation between each other.

[bollu]

MCMC is definitely not forgotten, it's still taught and used extensively by a bunch of folks, including those writing graphics engines!

[astro123]

I'm not suggesting that at all! I'm just saying it took 30 odd years to take off/break out of the nuclear physics world.

Though rereading my first comment that wasn't entirely clear...