(2019-01-17, 05:41 PM)malf Wrote: [ -> ]In terms of competing hypotheses, one must be that, once presented with colossal amounts of noisy data, mathematics can always produce some statistical significance.
If Bancel’s comments can be trusted we know the one thing the GCP is not measuring is GC.
It might be helpful if others suggested their favoured hypotheses in the next few posts?
I've been thinking about this in relation to Gelman's and Loken's "Garden of Forking Paths".
http://www.stat.columbia.edu/~gelman/res...acking.pdf
They refer to 4 different scenarios:
1. There is no flexibility.
2. There is flexibility, but no use can be made of it because of pre-registration.
3. There is flexibility and use can be made of it, but there is no fishing.
4. There is flexibility and there is fishing.
They focus on #3 reported as though it is #2. The assumption is that there will be an inflation of falsely significant results under #3, but not under #2. This parallels the claim made by Nelson, et. al.
This leads to two questions with respect to the GCP:
Are they acting under #2 vs. #3?
Is there no difference in the number of falsely significant results under #2 vs. #1?
Bancel's and other analysis shows us that there is a difference between the results you would obtain under #1 and the results of the GCP. That is, there is a difference in the results obtained when flexibility is present and when it is absent. For example, while there was flexibility at the beginning of the process in how the New Year's Eve data would be blocked and analyzed, for a number of years there has been no flexibility, as it is necessarily analyzed the same way each time. And the cumulative results are definitively non-significant (
http://noosphere.princeton.edu/events/newyear.2015.html). Bancel (
https://www.researchgate.net/publication...xploration) also showed that the choices made in the presence of flexibility, but without the purported ability to use that flexibility (i.e. "formal hypotheses"), among choices of which identical events to include, test statistics, and blocking intervals gave a different result than that expected if flexibility truly could not be used. And he also demonstrated a correlation between the measured correlations and the timestamp errors, whereas there should be no correlation in the setting of random error unless there is a process which selects for fortuitous timestamp errors.
So either they are really acting under #3, or there is an interesting phenomenon where there are differences between #2 vs. #1. Probably most non-proponents just assume that they are acting under #3, given that there isn't anything to prevent it. As you mention in a later post, there are ways to tighten this up. It would be interesting to see what happens if they do that. If the effect remains, it still won't be Global Consciousness, but it would seem to be anomalous in some way.
Linda