Lucia takes on the Gergis et al hockey stick screening fallacy

This is an excerpt of a larger post that deserves the wider attention WUWT can bring to it. The image at left is from my article: A look at treemometers and tree ring growth – Anthony

Screening: Now with good+bad treenometers.

Written by: Lucia Liljegren

In my recent posts, I discussed how screening treenometers (or any proxy) based on correlation with temperature during the calibration period can screwup bias a proxy reconstruction in ways that deceive the analyst who is either unaware of or refuses to believe in the types of biases that screening introduces. With respect to blog spats about AGW, the short unnuanced interpretations of the two limits of what screening can do are:

• “If you screen, a batch of treenometers that do contain some signal, you can exaggerate a hockey stick.”
• “If you screen, treenometers that contain no signal at all, you can create a hockey stick from trendless data.” and

Both issues clearly make it difficult to decree that any hockey stick contained in a reconstruction where the specific proxies were selected by screening with correlation.

(Note: What I mean by screening here is “peeking” at the data to decided what to toss out. It’s ok to believe certain conditions result in high correlation, decree you will go out and collect trees from pre-designated sites, and then use all tree cores from all sites. What you must not do is collect the trees and afterwards toss out any tree cores or sites based on correlation with temperature during the calibration period.)

Naturally, based on my two extremes, people whose intuition says there must be some advantage to screening want to know two things:

• Suppose you pick sites, and you were somewhat successful, but some of the treenometer were temperature sensitive and others weren’t.
• Couldn’t you figure out how to “improve” results from a batch of “good” treenometer and bad ones?

Today, I’m going to what your results would look like in two cases. In one, by picking sites, you got batch of “quite good treenometers” mixed with a batch of “not treenometers at all” and in another, you got “adequate treenometers” mixed in with “not treenomters at all”

Read the post here in entirety, well worth your time. The results speak for themselves.