Mark

#68 mosh, the trick is this: causality and correlation degrade over space at different rates for different reasons. Causality is always stronger locally than at a distance. Everyone knows that. Correlation, however, is a sample statistic – a statistic that is prone to error from many sources. So sometimes it is higher further away than closer. Correlation does not equate with causation; it merely samples it.

“Teleconnection” is a sensible term in climatology, where there is no causal inference in the way to bugger up its meaning; Pacific air masses are teleconnected with California weather. In paleoclimatology, however, you have the cause-and-effect response of the proxy entering into the equation. In my view paleclimatologists are just being too cute when they use the term. I am glad that Ed Cook sees fit to put quote marks around it when he uses it in his papers. Why does he do that? I don’t know. Maybe he thinks it’s being a little too cute, or maybe worse, misleading.

Ken, I think the paradox you want resolved is explained by causation being a high-frequency phenomenon and teleconnective correlation being a low-frequency phenomenon.

Bender, I agree that your explanation (excerpted in part above) is the only one that makes sense vis a vis a correlation with a remote temperature and not the local one, but I wanted JEG to do the explaining and give some details that explain the a priors. I have a major problem attempting to determine how much some of these climate scientists appreciate the dangers of data snooping. JEG and Rob Wilson come to mind. Upwelling teleconnections being, I would think, much more amenable to straight forward a prior explanation than the no-correlation-to-local-temperature teleconnections, would provide considerably less opportunity for data snooping. In either case one needs to provide evidence for a reasonable physical connection that can be calibrated, validated, and give expectations of being more or less permanent over the time period of the reconstruction.

I think that to which Steven Mosher is referring in his posts here is that on face value it appears that temperature reconstructionists look to correlate with local temperatures, but failing that, take a shot at correlating with a global or a teleconnected regionally remote temperature without a sound connecting principle given. I would like to see more time spent on the connecting principles.

I have witnessed some lengthy debates at stock investing web sites where it was my judgment that some of the more intelligent participants were the most difficult to convince of the dangers of data snooping because it appeared they were convinced they could outsmart the market and thus their investing models, no matter how much they lacked out-of-sample testing and exhibited over fitting, had to be merely confirming what they already had surmised as being true.

Was thinking of commenting earlier on this topic. Glad it’s come up.

If A is forcibly teleconnected with B is forcibly teleconnected with C, then at some point the teleconnections have to have global reach and C will reconnect back with A; it is a global circulation, after all. Are climatologists making a logical error assuming there is a pacemaker at the source end of the teleconnection? What if the “pacemaker” itself is being forced by weak inputs from X,Y,Z? For example, the Cook et al paper, where US drought is teleconnected back to the eastern tropical Pacific. If that area is running warm, that heat came from somwhere. But what if it comes from many places? Where does the search for teleconnections end? (Think about that texas sharpshooter. When does he get to go to the barn to draw his a posteriori targets?) And what if the X,Y,Z sources are always changing over time? How robust are those teleconnective correlations? Seems critical to me if you want to forecast the way Mann is.

Re: Ken’s quest for JEG’s lecture: I think the statistics of a global hunt for correlative telecennections is fraught with overfitting hazard (texas sharpshooter problem). JEG can probably explain the principle of teleconnection. I don’t think he can justify these correlations as anything but descriptive and highly optimistic.

Ken, I think the paradox you want resolved is explained by causation being a high-frequency phenomenon and teleconnective correlation being a low-frequency phenomenon. As long as the proxy models are mis-sepcified (e.g. conflating drought with temperature/growing season length), the low-frequency correlation at a distance might be higher than the local high-frequency correlation. The teleconnective hypothesis is nevertheless valuable because, given noisy data, the high correlation with the teleconnected pacemaker might prevent you from falsely concluding there is NO relationship between the proxy and its environment, when there may be a weak one. That doesn’t mean the teleconnective correlation is a correct description of the local response; it just means you need better data and a better model to obtain the correct response. If you did that you would see a higher correlation locally than abroad. Paradox resolved.

Is that evidence available for the upwellings that are described in this thread?

I would like to obtain more background information on teleconnected temperature proxies as it would appear the lesson I was expecting to obtain from JEG is not going to happen, or at least any time soon. In view of a teleconnection from an upwelling being consistent with the local temperature, that type of teleconnection is much easier to understand than one where temperatures in a local proxy are teleconnected/correlated to a global or other regional temperature while not correlating with the local temperatures.