Waxman Malarkey 3: Impact Zone Alaska

First, Fig. 1 clearly shows the dependence of Alaska temperatures on the Pacific Decadal Oscillation (PDO).

Spilled my coffee over this one!!
Fig. 1 is your picture of Baked Alaska. This seems not to have been picked up by previous commenters.
In any case, this figure, as a pictorial rendition of climate trends in Alaska, is about as accurate as the Waxman Malarkey.

Willis Eschenbach says:
July 1, 2010 at 2:07 pm

The crazy thing about language is that when a mistake becomes common enough … it becomes standard English, and it is no longer a mistake. That’s how the language changes.

Oh, I can’t resist, even though I may be totally wrong.
In the aftermath of Hurricane Katrina, several reporters referred to the decimation of the area. Decimate, I believe according to Richard Lederer, referred to a Roman Army practice of encouraging troops who hadn’t performed their best by killing every 10th soldier. That would provide the impetus to the remainder to offer more than 111% in the next battle.
Decimation seems to have spread from New Orleans to other areas and has also remained in Louisiana – it seems the oil spill is decimating the area now.
Hmm, I searched for Richard Lederer decimate and got http://www.wikihow.com/Use-Commonly-Misused-Words which says in part
Oftentimes, “decimate” is misused to mean “devastate,” which means “overwhelm or lay waste to.” Hurricane Katrina devastated New Orleans, but did not decimate its population.
Yay. Decimate Global Warming.

In re to font, fount, front, etc (and more than slightly OT):
The American language, mostly an enlarged set of its principal precursor…the English language…is a dynamically growing language which for better or worse is constantly evolving. Words which may have held a single principal meaning or preferred pronunciation a Century ago have diferent meanings today. A brief summary of the problem this presents to lexicographers can be found in the definition of “irregardless” in the Eleventh Edition of the i>Merriam Webster Collegiate Dictionary which publisher I consider to be the most authorative source for the contemporary American language.
In my middle years I was very resistant to even subtle changes in what was at the time the “accepted” definition (or pronunciation) of words. In recent decades I’ve grown more tolerant in accepting these changes…abrupt or nuanced…which leads to the irony: an accepted definition 0r pronunciation reigns supreme, until a majority of the population thinks otherwise. There is nothing so scorned by the elite as a formerly
“correct” word which has been redefined or replaced by common usage.

@Willis Eschenbach,
Bad manners on my part shouldn’t be an excuse for not giving a complete explanation of the numerical disagreement in our estimates.
As for autocorrelation, what model did you use that gave you your results? The correlograms I plotted for the time series showed no lags that were significant, with the exception of a marginally significant 8 year lag for the annual series but not the winter series.
When I added the 8 year lagged temperature to the annual models, the yearly trend for 1895-2009 was unchanged at 0.007 °F / year (p = 0.04). For the 1970-2009 period, the trend attributable to year alone decreased to 0.03 °F / year (p = 0.01), and this is clearly a conservative estimate since the lag term absorbs some of the trend. For the winter data, there’s no significant autocorrelation, so my results are unchanged.
As before, I’m making all of my work public, and I’m asking that you do the same.

@Willis Eschenbach,

Try the method of either Quenouille or Nychka for dealing with autocorrelation, that’s what I use.

Thanks for your kind and to-the-point response.
I didn’t think to correct for AR1 non-independence in the residuals because the data do not suggest that there is any, with the exception of the weak signal at 8 year lag in one of the two series.
With your mention of Nychka, are you suggesting that the corrected sample size

n_e= n * [1 – rhohat – (0.068/sqrt(n))] / [1 + rhohat +(0.068/sqrt(n))]

be used? When I run that analysis, I get significant trends for the annual means, both full (p = 0.05) and 1970-2009 (p = 0.01) and trends that are not significant for the winter means, both full (p = 0.07) and 1970-2009 (p = 0.09). The p-values you give for these trends are p = 0.06, not given, p = 0.06, and p = 0.12, respectively. I still don’t know why are figures are in disagreement.
From Nychka et al. (2000):

However, to date a correct treatment using maximum likelihood or Bayesian approaches requires specialized statistical software […] By comparing this approach to the more sophisticated maximum likelihood estimators we find negligible difference.

In 2010, using (the freely available) R, the MLE approach is much easier than kludging around with Nychka. Directly adding an AR1 term, I get p = 0.04, p = 0.02, p = 0.06, and p = 0.06 for the four time series. This agrees with the admission in Nychka et al. (2000) that “when the ADJUSTED effective sample size is large, the coverages may become somewhat conservative.” This is the case with this data, where the number of samples is large and the estimated lag-1 autocorrelation is low.
tl;dr Thanks for citations. Annual data may have autocorrelation but it doesn’t matter. Winter data doesn’t have autocorrelation but assuming it does weakens the tests just enough to miss significance. Our numbers still disagree. I’ve shown where mine come from. You haven’t. Your claim that “there is no statistically significant trend anywhere” is falsified.

@DirkH,
Please don’t derail.
I can try to fix up anything I didn’t explain well enough, if you’d point it out.