McShane & Wyner Hockey Stick Smackdown – redux

Response up at RC.

I belive that this exchange exposes a vital part of the teams paleo effort.
From the team:
“In the frozen 1000 AD network of 95 30 proxy records used by MW, 36 tree-ring records were not used by M08 due to their 31 failure to meet objective standards of reliability. These records did not meet the minimal 32 replication requirement of at least 8 independent contributing tree cores (as described in 33 the Supplemental Information of M08).”
Reply by MW:
“The process by which the complete set of 95/93 proxies is reduced to
59/57/55 is only suggestively described in an online supplement to Mann
et al. (2008)3. As statisticians we can only be sceptical of such improvisation,
especially since the instrumental calibration period contains very few independent degrees of freedom. Consequently, the application of ad hoc methods to screen and exclude data increases model uncertainty in ways that are unmeasurable and uncorrectable.”
If you have a preconceived notion as to what you intend to find in your data (in this case unprecedented warming during the last 50 years) and than look for data that fits that description (in the teams words “sensors that are able to pick up the warming”) you will naturally end up in a very biased and potentially very wrong place.

Here’s the bit I like from the M&W reply:
“The fact that our paper was of interest not only to academic statisticians and climate scientists but also economists and popular bloggers [1] bespeaks of the importance of the topic.”
and Footnote [1] says
“Steve McIntyre of Climateaudit and Gavin Schmidt of Realclimate“. 🙂

Sam the Skeptic says:
December 13, 2010 at 10:39 am
————
And he surprises me by falling into such an obvious trap.
The concentration of CO2 has increased (in round figures) from 300ppm to 400ppm – in other word from 0.03% to 0.04%
————
It does not surprise me that you would fall into such an obvious trap.
A gas, in a mixture of gases, acts independently of all the other gases in the mixture. That is why chemists use the concept of partial pressure. Hence the 0.03% value. which is the proportion of CO2 relative the other gases, is not relevant to the ability of CO2 to absorb IR.
The partial pressure is important and that increased by the ratio 4/3.
Hope that’s clear now.
If you don’t understand trivia like this maybe you should keep you mouth closed instead of slagging off at people who do understand the basics.
[Back off the attacks of other posters. Stick to the issues / science. Hope THAT is clear now… bl57~mod]

HAS says:
December 13, 2010 at 11:25 pm
……….
“The thing that worries me is that putting aside the detail of what the data does or does not show under various analyses M&S don’t seem to address basic issues raised in MW’s rebuttal. Did M08 quality control introduce bias; exactly what assumptions were required to lead to MW suggesting the recent warming was anomalous in their original paper; were MW correct to say that the literature suggests more than 4 principal components could be acceptable; did or did not MW test other methods in addition to Lasso as part of their rebuttal; did or did not MW properly address the criticism by other commentators about other naive models eg composite regression; how can M&S refer to “pseudoproxy” experiments with a straight face without first dealing with MW’s critque of these?”
++++++++++++++++++++++++++++++++++++++++++++++++++++++
You, uhmm, read the discussion papers and the rejoinder? From SMR ….36 tree-ring records were not used by M08 due to their 31 failure to meet objective standards of reliability.” Here, they admit to introducing bias to M08.
You asked, “Did M08 quality control introduce bias; exactly what assumptions were required to lead to MW suggesting the recent warming was anomalous in their original paper;”[?]
Your not reading the original paper properly. They created a model they thought would best recreate temps from the proxies. Later in the paper, they assert the models (any and all) can’t recreate temps from the proxies. In other words, they don’t believe their model to be accurate.
Lastly, you ask, “did or did not MW test other methods in addition to Lasso as part of their rebuttal; did or did not MW properly address the criticism by other commentators about other naive models eg composite regression; how can M&S refer to “pseudoproxy” experiments with a straight face without first dealing with MW’s critque of these?”
From MW rejoinder……“We are able to show, by brute force computation, that our results are invariant
to these choices. Furthermore, as stated in our paper, we implemented
many of these proposals prior to submission (for discussion of variations
originally considered and justification of our choices, see Section 3.7 for
the Lasso; footnote 8 for thirty year blocks; Section 3.4 for interpolation;
and Section 3.6 for calibration to local temperatures). In contrast, we credit
McIntyre and McKitrick (MM) for pointing out the robustness of these results
and Kaplan for actually demonstrating it by using Ridge regression
in place of the Lasso (see Kaplan Figures 1 and 2). We direct the reader to
our SI where we perform the same tests (1) for a plethora of methods (including
the elastic net called for by HU and the Noncentral Lasso called for
by Tingley) (2) using thirty and sixty year holdout blocks (3) using both interpolated
and extrapolated blocks and (4) fitting to the local temperature
grid as well as CRU when feasible. Once again, the results demonstrated
by Figures 9 and 10 of our paper are robust to all of these variations.
Try reading the thing instead of listening to talking points from people that don’t understand what they are reading. Thanks

From MW2010 Rejoinder,
“Before embarking on our discussion of their work, we must mention that,
of the five discussants who performed analyses (DL, Kaplan, SMR, Smerdon,
and Tingley), SMR was the only one who provided an incomplete and
generally unusable repository of data and code.”

————–
SMR have not learned the openness lesson yet. But there is hope for SMR, which is that M & W are good teachers.
John

lol, so I’m reading RC’s response to the rejoinder. Either they didn’t read the paper and rejoinder, or they do not understand what they’re reading. Sorry for the length, but I’m procrastinating other tasks.
The first criticism, in the paragraph that starts, “On that specific issue, presumably just an oversight,……” Here, RC is speaking about the differences in the proxies used. MW used the entire data set while M08 used a subset, removing some from the database.(Again from RC) ..”….This is even more true when the frequently challenged “Tiljander” series are removed, leaving a network of 55 series….”
So, what did MW say to this? Well, they said plenty! In the interest of brevity, I’ll only reproduce a few statements, the first, being the most important, in my estimation. From MW rejoinder…..”The process by which the complete set of 95/93 proxies is reduced to 59/57/55 is only suggestively described in an online supplement to Mann et al. (2008)3. As statisticians we can only be skeptical of such improvisation, especially since the instrumental calibration period contains very few
independent degrees of freedom.” Also, from the comments of figure1….”Second,
the RegEM EIV methods produces reconstructions which are nearly identical to those produced by OLS PC4 and OLS G5 PC5. Compare also with SMR Figure S2 which is similar to the bottom panel but excludes RegEM EIV.” Again, from the rejoinder…..”The appearance of a difference in SMR Figure 1a is especially magnified because those reconstructions are smoothed. Smoothing exaggerates the difference and requires careful adjustment of fit statistics such as standard errors,….” Well no wonder they don’t look the same. It has nothing to do with including or excluding data. What happens when their methods and data are compared to M08?…….Well, RC could have just looked at figure 2 and then read underneath it…….again, from the MW rejoinder under figure 2 (bold emphasis mine)…..”Difference Between our Bayesian AR2+PC10 Model of Section 5 and Various Other Models. The left panel gives the difference between our Bayesian AR2+PC10 model fit to the network of 93 proxies dating back to 1000 AD and the original Mann et al. (2008) RegEM EIV fit to the network of 59 proxies dating back to 1000 AD. The right panel gives the difference between our Bayesian AR2+PC10 model fit to the network of 93 proxies dating back to 1000 AD and the model of SMR Figure 1b (i.e., the same Bayesian AR2+PC10 model but fit to the network of 55 proxies dating back to 1000 AD instead of the network of 93 proxies). As can be seen, there are no statistically significant differences between these two models and our Bayesian AR2+PC10 model fit to the 93 proxies.
This, in itself and by itself should have satisfied most of their critiques. BTW, as noted by MW, the tree rings had already undergone a series of minimal standards. As noted…….”(ITRDB, version 5.03: http://www.ncdc.noaa.gov/paleo/treering.html). All ITRDB treering proxy series were required to pass a series of minimum standards to be included in the network: (i) series must cover at least the interval 1750 to 1970, (ii) correlation between individual cores for a given site must be 0.50 for this period, (iii) there must be at least eight samples during the screened period 1800-1960 and for every year used.”
That isn’t to say RC doesn’t have a sense of humor about all of this, from RC(again, bold is mine), “MW claim that M08 quality control is simply an ‘ad hoc’ filtering and deny that they made a mistake at all. This is not really credible, and it would have done them much credit to simply accept this criticism.”—–there isn’t much to say on this, one can only laugh. Thanks RC!!
Right after that, they kick in with their 80 % anomalous finding of MW, stating this was a conclusion of MW. Oddly, that number isn’t in MW’s summation or conclusions. This is an example of RC being willfully dense. Talk about misrepresenting someone else’ work! From the original MW10 “Since our model cannot detect the recent temperature change, detection of dramatic changes hundreds of years ago seems out of the question.” if that weren’t enough, from the rejoinder,(bold mine) “That is, we lack statistical evidence that the recently observed rapid rise in temperature is historically anomalous.” Does this sound like MW put any stock in the results of their model? I think RC is being disingenuous when using the 80% number. They can’t be that dumb, can they?
Then they speak of PC’s (principle components) they state PC10 is too large and advocate using PC4. All RC had to do, was to look at figure 4 in the rejoinder. Again, here’s what MW rejoinder stated,(bold mine) “in Figure 4 (we include an unsmoothed plot in the SI; since SMR prefer four principal components, we include plots for a Bayesian AR2+PC4 model in the SI and note that the four and ten PC models perform almost identically). In other words, PC4 vs PC10 makes little or no difference.
Then RC goes on about IPCC standards and levels of confidence. I can only suppose that RC believes normal science is done in the same manner as climate science such as 50% being 90% confident that they’re engaged in real science.
The rest seems to be hand waving and cheer leading for other critiques and a snark towards M&M with the exception of this statement regarding centering…..”All the PC calculations use prcomp(proxy, center=TRUE, scale=TRUE) to specifically deal with that, while the plots use a constant baseline of 1900-1980 for consistency.”…..I wish I had a better grasp of the vernacular of both climate scientists and statisticians, but I really don’t believe MW were speaking toward a programming command.
Honestly, my take is that RC all but entirely ignored the rejoinder.
Last comment from RC…….”In summary, there is much sense in these contributions, and Berliner’s last paragraph sums this up nicely: ‘ The problem of anthropogenic climate change cannot be settled by a purely statistical argument’………“————— wow, most agree with this.

Stein, in his editorial says,

An example of an implausible sharp null hypothesis would be that a large increase in the concentration of CO2 in the atmosphere has exactly zero effect on the global mean temperature.

It may not be the political correctness of the day, but I have to say this:
While it is currently accepted that a doubling of CO2 will raise the global average by 1C, because of the effect CO2 has on escaping heat energy in the IR range, that is based on the theoretical. Many, MANY lab and theoretical concepts do not hold true in the empirical world – the real world. Lab tests in biology and physics are not accepted by themselves, but must be measurably verified in the field.
Until that is done, all the formulas in the world won’t convince me that something that makes up 1/26th of 1% can have a measurable effect on the climate. Am I leaning toward accepting it? Yes. But I lean toward a lot of things that may or may not be actually true in the real empirical world, so the weight of my leaning doesn’t mean anything.
But what is meant by “large increase”? The 100% increase giving a 1C – that is very very large increase. Everyone ASSUMES that a 100% rise is plausible. The increase since the mid-1800s is about 100 ppm, about a 36% rise. There has been a 0.7C rise since 1900. The rise is ASSUMED to connect the two as cause and effect. I think that cause-effect connection is tenuous at best. The degree of change seems close, but that may be a coincidence. I am not convinced. That 36% rise in CO2 may or may not be connected to the 0.7C rise in temperature.
Especially the people who suspect that the rise in global temperature is an artifact of the CRU methodology (particularly in UHI adjustment and arbitrary TOB adjustments), why do they accept this claim as valid? Take away the CRU adjustments and just how much rise in temperature have there been? We simply don’t know.
Since we don’t know what the rise is, the lab or theoretical cannot be accepted at face value.

>While drought may develop locally, or rainfall or flooding elsewhere (as you state) the extra heat is not, as you assert, resulting in extra evaporation locally.
Huh? Look at the Clausius-Clapeyron equation for saturation vapor pressure, as that is what is relevant: de_s/dT = Lv*e_s / R_v*T^2
So how do you claim the factors of T^2 and e_s are not important?????