Bancel Remarks, GCP Critique Thread
Date: Fri, 12 Oct 2007 09:37:26 +0200
I'm happy reading these posts, because I feel that there is general agreement on many of the issues.
Regarding distance dependence, York states correctly that no specific distance dependence is required, but that *some* distance dependence is implied by the notion of field reg; alternatively, an infinite uniform effect could still be imagined as a "field", but a trivial one (and uninteresting one in the sense it would not lead to further hypotheses about distance). James seems to agree in saying a 1/r^2 was mentioned simply as an example. York's "best guess" of a drop off at distances beyond the earth's size is a way of thinking about field-reg in the absence of data sensitive enough to probe shorter distances.
Regarding the general GCP hypothesis, Roger is completely correct that the formal experiment tests a valid hypothesis. The issue - which I didn't explicitly go into in an earlier post - is that the hypothesis is so broad that it doesn't tell us much beyond simply demonstrating that something is going on in the data. I have always felt that this generality of the GCP hypotheses is a brilliant approach. I frequently come back to a quote from Anthony Freeman in a discussion of consciousness studies. Alluding to John Searle, he helpfully lays out the obvious when dealing with research in a nacent field:
"The importance of not overdefining the topic ahead of investigation is routinely stressed by John Searle... He distinguishes between an analytic definition, which is arrived at only at the end of an investigation, and a commonsense definition, which comes at the start and serves to identify the target of the research program. He advises taking as a starting point a simple statement such as this one: Consciousness consists in the those states of awareness that begin in the morning when we wake from dreamless sleep and continue throughout the day until we fall asleep again, or fall into a coma or die."
It's easy to see how this carries over to the GCP hypothesis.
As Roger says, he wisely makes a conscious effort to keep the hypothesis broad, even including events based on "anti-hunches". In statistical terms, the GCP hypothesis is a complex hyothesis, as opposed to a simple hypotheses which would have more interpretive power, but would possibly not get us very far, given our ignorance at this stage.
Regarding fishing, the formal experiment is thus not fishing at all. Further, aside from providing a demonstration of some there there, the formal experiment delimits where in the mass of data one might look further. My role in this happy collaboration has been to focus on that. The logical guess is that there should be additional anomalous structure (temporal, spatial, etc) in the subset of data corresponding to events. And that is what the analysis (we can drop calling it fishing) is starting to reveal.
Regarding 9/11, that particular day was so unusual, both as an event and in terms of how the data behaved, that it is difficult to interpret. It remains an outlier in several aspects. Roger is right to emphsize the importance of finding a "data" outlier precisely where we have a singularly unusual event. On the other hand, it is diffucult to understand in the context of what we have found in the analyses, because the data deviations for 9/11 are somewhat different than what we characteristicly see in the event data. I think that the 9/11 data holds some secrets, but I don't yet see how to unlock them. We will certainly return to the 9/11 data as we learn more.
best to all,