5 ms·

Looks like bogus statistics to me. (Caveat: I only read the sciencealert link not the underlying publication). But with a sample of 89 "events" over a purport

by dzdt 5y ago

Looks like bogus statistics to me. (Caveat: I only read the sciencealert link not the underlying publication). But with a sample of 89 "events" over a purported 10-cycle period we are in the small sample realm where it is easy to get spurious apparent relationships. People love to see patterns, and if you aren't really careful you can "prove" a pattern in essentially random data.

wrycoder 5y ago
Underlying: https://www.sciencedirect.com/science/article/pii/S1674987121001092?via%3Dihub https://www.sciencedirect.com/science/article/pii/S167498712...
- chris_va 5y ago
  Section 3.3 there, the Fourier analysis, looks really weak. If you just plot 89 random numbers between 0 to 270 it'll have strong spectral peaks. Their method: > After permuting the order of the intervals, a new time series was re-calculated based on the new set of intervals. This method ensures that the test time series are between 0 and 260 Ma and the number of events remains at 89. Ultimately, only ~ 4% of the test datasets produced a higher spectral power at 27.5 Myr, indicating a confidence level of at least 96% for that period. ... seems very fishy to me. I could believe that the spectral power at that frequency is the mode in the data. I don't see a proof that the particular spectral power associated with 27.5 is statistically significant given a random set of events.
  - dzdt 5y ago
    It is incorrect to take 27.5 Myr as a prior fixed quantity when there was no prior theoretical reason to expect that period. Correct analysis would ask, given a randomly generated data set, what are the odds as strong a signal as this is generated for any recurring period in some range (like 3Myr-100Myr)
nosianu 5y ago
If you had read the article you would have been informed that this publication is one in a long line of similar ones going back as far as 1920 who all came to similar numbers. You are right, if there was only this one paper it would be entirely unremarkable and informing the general public would be wrong. But that is not the case because the article fortunately does not limit itself to pointing to just that publication.
- SeanLuke 5y ago
  Just because everyone's seeing the same pattern in random data doesn't mean there's a pattern at all. For very good evolutionary reasons, humans are extraordinarily good pattern-recognition machines: we see ghosts in data when nothing exists. Our collective false positives are quite high.
- CRConrad 5y ago
  Going back to about 1920, everyone saw canals on Mars, too.

by dzdt 5y ago

wrycoder 5y ago

Underlying: https://www.sciencedirect.com/science/article/pii/S1674987121001092?via%3Dihub https://www.sciencedirect.com/science/article/pii/S167498712...

chris_va 5y ago
Section 3.3 there, the Fourier analysis, looks really weak. If you just plot 89 random numbers between 0 to 270 it'll have strong spectral peaks. Their method: > After permuting the order of the intervals, a new time series was re-calculated based on the new set of intervals. This method ensures that the test time series are between 0 and 260 Ma and the number of events remains at 89. Ultimately, only ~ 4% of the test datasets produced a higher spectral power at 27.5 Myr, indicating a confidence level of at least 96% for that period. ... seems very fishy to me. I could believe that the spectral power at that frequency is the mode in the data. I don't see a proof that the particular spectral power associated with 27.5 is statistically significant given a random set of events.
- dzdt 5y ago
  It is incorrect to take 27.5 Myr as a prior fixed quantity when there was no prior theoretical reason to expect that period. Correct analysis would ask, given a randomly generated data set, what are the odds as strong a signal as this is generated for any recurring period in some range (like 3Myr-100Myr)

nosianu 5y ago

If you had read the article you would have been informed that this publication is one in a long line of similar ones going back as far as 1920 who all came to similar numbers. You are right, if there was only this one paper it would be entirely unremarkable and informing the general public would be wrong. But that is not the case because the article fortunately does not limit itself to pointing to just that publication.

SeanLuke 5y ago
Just because everyone's seeing the same pattern in random data doesn't mean there's a pattern at all. For very good evolutionary reasons, humans are extraordinarily good pattern-recognition machines: we see ghosts in data when nothing exists. Our collective false positives are quite high.
CRConrad 5y ago
Going back to about 1920, everyone saw canals on Mars, too.