Tuesday, March 27, 2012

Selection Bias on Extreme Events

I meant to write more about extreme events but I haven't yet figured out exactly how to say what I want to say, so in the meantime here is an crucial paragraph from the recent Coumou and Rahmstorf paper (emphasis mine):
"There are many types of conceivable extreme, such as for different regional entities or time periods as well as different weather parameters (some 27 indices for extremes have been proposed). To pick the type of extreme post hoc — for example, to study Pakistan rainfall extremes after a record-breaking event there — risks selection bias, that is, bias by selecting just the kind of time series that shows recent extremes. Proper statistical analysis of changes in the observed number of extremes thus requires: (1) a single, comparable type of extreme; (2) selection of time series by a priori objective criteria; and (3) sufficiently long-running high-quality data."
This is important, and, I suspect, where many AGW believers like Bill McIbben start to go wrong.


Michael Tobis said...

Yes, this is a real problem. I call it the baseball problem. It's like the "most extra base hits by a left-handed rookie shortstop in a double-header away game" type of record. These sorts of things come up all the time.

But then are sorts of things that do not come up all the time. The recent extended and extraordinary warm period in the midwest and the 2010 event in Russia have many similarities and no precedents.

So selection bias is not the only trap. It is easy to avoid selection bias by refusing to do any attribution whatsoever, but you're guaranteed of false negatives if a real positive ever comes along.

In fact, there is hardly any room for doubt that some recent severe events were in fact exacerbated by anthropogenic climate change.

The time for asking this sort of question has passed. As evidence collects, what is a sound observation at one time may become unsound later.


This can be overstated, of course, but it is no longer the case that the connection can be broadly dismissed, as you imply.

It's not a maybe anymore. It's already happening.

David Appell said...

Michael, thanks for the comment.

I know that you and others keep saying that "there is hardly any room for doubt that some recent severe extreme events were in fact exacerbated by AGW," but where is the scientific proof of this?

I think the risk of selection bias is huge here.

I need to read Trenberth's new paper in detail, the one in the link you gave...

Anonymous said...

"-there is hardly any room for doubt that some recent severe extreme events were in fact exacerbated by AGW-, but where is the scientific proof of this?"

There is not proof, just a high statistical likely hood that the types of events described in the growing body of literature were more likely to happen and more likely to be exacerbated due to the change in atmosphere and change in energy balance.

This really should be good enough for people to pause and think deeply on the implications of this anomaly at this stage in temperature change. And we're blowing right on past this for years to come. If you are waiting for proof, or exact percentages for the attributions of extreme events in order to form a strong opinion, then you will continue to waver without ever actually understanding what is being discovered.