Nature it ain't

Last week I wrote about
how Bob Carter was out by a factor of 20 in an estimate of how much
warming could be attributed to human activity. He has now posted the
text of another
talk where he
gives a source for his bogus claim. It's href="http://www.foxnews.com/story/0,2933,123013,00.html"
rel="nofollow">this FOXNews opinion piece by Steve
Milloy
. Carter is a
Research Professor at James Cook University, so you would have thought
he would be aware that opinion columns by non-scientists aren't the
best source of scientific information, but I guess not.

Some highlights of his talk: He said:

Their assertion is a symptom of a disease called Hansenism which has
gripped western media sources and political, business and public
opinion in a deadly grasp. Hansenist climate hysteria is driven by
relentless, ideological, pseudo-scientific drivel, most of which
issues from green political activists and their supporters, and is
then promulgated by compliant media commentators who are innocent of
knowledge of true scientific method. Opportunistically, and sadly,
some scientists, too, contribute to the Hansenist alarmism. Sir
Roderick Carnegie was quite correct when he formerly identified such
environmental lobbying and emotional propaganda as a greater threat to
our society and way of life than, in its heyday, was communism.

James Hansen. Worse than Stalin and Mao COMBINED!!

Why Hansenism? Because James Hansen was the NASA-employed scientist who started the climate alarmism hare running on June 23, 1988, when he appeared before a United States Congressional hearing on climate change. On that occasion, Dr Hansen used a misleading graph to convince his listeners that warming was taking place at an accelerated rate (which, it being a scorching summer's day in Washington, a glance out of the window appeared to confirm).

What actually happened was
that Hansen presented to Congress a graph showing scenarios for high,
medium and low CO2 growth and said that the medium growth
scenario was most likely. The medium growth scenario has turned out
to be a good prediction of the subsequent increase in temperatures
When Michaels testified before Congress ten years later he
erased the medium and low curves and claimed that
because the high prediction was wrong, the climate model was faulty.

And while we are on the subject of misleading graphs, Carter presents
a
graph
that shows average temperatures falling since 1998. Oddly enough, he
uses a 25 month moving average to smooth the curve instead of the
conventional five year moving average. If you smooth it in the
normal way

the average doesn't fall, but increases steadily.

Carter goes on to say that "Hansenism" is like Lysenkoism, only worse,
cite Bray's bogus survey and the Oregon petition and to suggest that
Australia hire Bjorn Lomborg to run an Institute of Environmental
Assessment because CSIRO and BOM can't be trusted.

I dunno about the last one, maybe Philip Cooney would be a better choice?

I have rewarded Carter with his own category on my blog.

Tags
Categories

More like this

It must be taken as a good sign that septics use Hansen as backing for their assertions, yet there are others that use little namie-names to marginalize.

D


And while we are on the subject of misleading graphs, Carter presents a graph that shows average temperatures falling since 1998. Oddly enough, he uses a 25 month moving average to smooth the curve instead of the conventional five year moving average.

Well, there's not much point using a 5-year moving average if you only have 7 years of data.

That said, I agree with the 3rd comment RE 1998 looking very cherry-picked.

why exactly is the 5 year moving average the conventional thing?

5-year moving averages aren't writ in stone: however, a two-year average is unusually short (and that's what a 25-month moving average is--12 months before and 12 months after the month in question). I just finished calculating the 25-month, 37-month, 49-month, and 61-month (i.e., centered 2 year, 3 year, 4 year, and 5 year) moving averages: only if you cherry-pick both 25-months and 1998 do you get a "cooling trend."

And of course the same methodology applied to the seven years leading up to the 1998 high would show a rather stronger opposing trend with a sharp spike at the end. Speaking of hot years, the last I heard (April?) the 2005 trend was toward a record high year. I failed to bookmark and now cannot find whatever page it is that keeps track of this (I believe on a monthly basis), so would someone please post that link? Thanks.

By Steve Bloom (not verified) on 21 Jun 2005 #permalink

I followed the Fox/Milloy link and found that the scientific authority cited in turn by Milloy to back up his assertions was (drumroll)... Fred Singer!

By Steve Bloom (not verified) on 21 Jun 2005 #permalink

Moving averages as commonly applied, i.e. flat coefficients across all months used up to a 100% cutoff at the end of the period, are not a Good Thing. They are mathematically isomorphic with a resonant filter, which also produces a sharp cutoff just beyond the resonance peak; and, as such, they produce "ringing", a spurious peak at a particular frequency. Thus the anomalous behavior of the 25 month moving average versus other lengths of period. A preferred technique is the exponential moving average, where each term is reduced by a constant fraction with each iteration but never drops out; i.e., Jan/2+Feb/4+Mar/8+Apr/16+... etc. which has a more rounded shoulder for the cutoff, but does not exhibit any resonance.
Here endeth the lesson.

By z from kdlangland (not verified) on 21 Jun 2005 #permalink

Steve,

try here for Aus and here for global.

One certainly has to congratulate Bob on the sophistication (and sneakiness) of his arguments..... they are brilliant. His key points that most accumulated CO2 in the atmosphere does not come from human fossil fuel burning and that global temperature variability in the past is greater than that observed recently are indisputable facts. Unfortunately, they are also irrelevant.

1998 is hardly "cherry picked" - it's a data point. It would have been "cherry picked" if he'd started the graph at 1998. In fact, it started in 1970, which is fair in anyone's book as this is when the recent warming trend started.

Also, as was pointed out earlier, if you're trying to demonstrate a trend over seven years, there's not much point in using a five-year moving average.

In fact, at the end of the five-year average chart, you can see the average start to fall as the exceptionally warm 1998 year falls out of the average. That trend is likely to continue for a year of two unless 2005/06 prove to be exceptionally warm.

Not really a great example of "chartmanship".

1998 is hardly 'cherry picked' - it's a data point. It would have been 'cherry picked' if he'd started the graph at 1998.

Actually, it is cherry-picked, because he chose the starting year for his trend: if he'd used 1997 or 1999 the trend wouldn't've gone his way. Here is a graph that shows the yearly mean (unsmoothed) compared with the 3-year, 5-year, 7-year, and 9-year running means. In order to make the statement "cooling trend," you either have to do something odd like a 25-month mean, or you have to use a 7-year running mean and accept that your cooling trend is only one year long. Not much of a trend.

Running means like these are always "centered," so a 3-year mean is the year before, this year, and the next year. That means that the span of a smooth is almost always an odd number of periods, like 3, 5, 7, 9, or 11 years. That's why a 5-year running mean over the seven years from 1998 to now isn't as nutty as one might think.

As for the 25-month running mean, I should have emphasized how odd this is in my previous post. There is huge serial correlation across the months of each year, so a 25-month running mean really just takes the 12 months before, this month, and the next 12 months. This is almost guaranteed to make an oddball year like 1998 stick out, making it possible to talk about a cooling trend since 1998. (This is also why it's unnecessary to worry too much about using a fancier exponential smoother over months, as Z from kdlangland suggested; the serial correlation is too strong, and why everyone does a uniform mean across years, and then does running means of the yearly averages. Note, however, that Carter uses a Gaussian kernel for his smooth, which will also tend to do less smoothing of the 1998 values).

All of the above is meant to show how difficult it is to use these data to be able to make a "cooling trend" statement: you have to choose an unusual way to smooth and then choose the starting year over which to measure the trend. That's cherry-picking.

Actually, the five year average dipped slightly when 1998 fell out of the average, but since then it has increased and is now higher than it has ever been (as are the 10 and 25 year moving averages).

See this for Carter's cherry-picking of 1998.

Thanks for the graphs, Robert.

I might start by making it clear that I'm not defending Carter's conclusions. Seven years since the 'outlier' of 1998 is too few to be judging a trend. If 2005 & 2006 proved to be cooler than 2004 then one might have to think about it. And in fact, he's wrong anyway about a cooling trend " I would describe it as 'flat'

My interest, as a statistician, is only in the graph, and whether it is a fair representation.

I can't see that 1998 is in any way 'cherry-picked'. He's saying that temperatures since 1998 have not been as high. That's simply a fact. It's like saying that the peak of the tech-boom has been cherry-picked because stock prices fell after that.

Robert shifts the ground to argue that the 25 month moving average is 'cherry picking'. I have no idea why Carter describes it as '25 months' (maybe it's something to do with the way the calculation is done). But it's ridiculous to suggest that 24 months or 26 months would make any practical difference.

Personally. I think a longer averaging period would be desirable. Nevertheless, in the charts Robert presents, the 7 year trend dips, as would the five year if the next year was included (as per Tim's original post).

I'll conclude by saying that I don't think Carter's graph is misleading, not do I think it's 'cherry picked'. That said, it's pretty obvious from the raw data that the trend is upward, and assuming 2005 & 2006 are warmer than 2004, that's likely to be evident whatever scale of moving average you care to use.

Tim says that the five year average is now 'higher than it has ever been' " do you have a chart for that, Tim?

James wondered:

Robert shifts the ground to argue that the 25 month moving average is 'cherry picking'. I have no idea why Carter describes it as '25 months'

I'm not the one who shifted the ground: Carter describes it as 25 months because he actually used the monthly temperature averages rather than the yearly temperature averages that almost everyone else uses.

Tim says that the five year average is now 'higher than it has ever been' " do you have a chart for that, Tim?

That's the third panel of my graph. You can see that the 5-year running mean dipped right after 1998 and then has come back up, as Tim described. The 7-year endpoint dips because 1999 and 2000 were cooler than 1998 -- you can see that from the top panel -- but you can also see that the single-year average bounced back up after that. My graph includes the data through April 2005. In these data there's strong seasonality (as you'd expect) but the Jan-Apr mean isn't a terrible estimator of the full-year mean -- nonetheless, it's too early to say whether 2005 will end up being warmer, cooler, or the same as 2004.

Robert,

While I agree with everything you've posted subsequently, you did call the use of 1998 as "cherry picking" (see post #3).

While not defending Carter, there is nothing wrong with the graph he uses. It's up to him to use the data in a way to make his case . I would be critical if he used different criteria in the future.

James:

I appreciate your comments, and do not mean to be argumentative -- though I can sorta see that it may come across that way.

Since you're a statistician, here's a little project: take the data that Carter cites.
The data are monthly means, with the last column being the yearly mean. Every other line is the number of stations reporting each month. If you decide to work with the monthly (rather than yearly average) data, note that there is seasonality so you might want to think for a moment about the width of any smoothing span.

Now here's the question: **how many ways can you smooth either the monthly or yearly data in order to make the statement "cooling trend from year X to now" be true? **

If you do this you'll get a sense of how easy or troublesome it is. I would argue that the more effort it takes to make the statement be true, the harder it is to believe that Carter did not cherry-pick. Here's a confession that I'm afraid reveals me for the anal-retentive geek that I am: I did all of this before I posted the very short message #3 above.

1998 is a data point but it's a data point with a big noise hit, el nino not being just deterministically related to global warming. When you have a big noise hit, you need a lot of data points to overcome the contaminating effect of that noise hit. Using just 2 data points as in a 2 year (approx) average is not going to get rid of the contaminating effect of the 1998 el nino.

If you use every data point since and including 1998 in a regression then even though the first point is contaminated you still actually get a +0.01 degrees C/ year slope. So even with this contamination there is actually an overall warming trend since 1998. In fact it's still positive even from 1998 to just 2003, i.e. six data points.

So to my understanding, anyone who says there has been a cooling trend since 1998 is talking rubbish. Not even when you ignore data contamination can you justify saying there has been a cooling trend. 1998 might have been unusually warm, but the two immediate following years 1999 and 2000 were much cooler. A regression over the 6 or 7 years takes this into account.

By Chris O'Neill (not verified) on 22 Jun 2005 #permalink

Re #14, if an economist were to take the six months or so after the bottom fell out of the tech stocks and use them to project a multi-year trend (i.e., a continued sharp decline) without considering other factors, the result would fail the laugh test. Carter appears to have done the same thing. If that's not cherry-picking, what would cherry-picking be?

Re #10, thanks for the link (which I have bookmarked).

By Steve Bloom (not verified) on 22 Jun 2005 #permalink

Tim,

Hansen's curve does not show variation of mean temperature over time, but the variation of the variation around the mean temperature over time.

When a temperature anomaly is computed, it is the reading's departure from the running mean, (30 year moving average) that is being graphed. Not the variation of the earth's mean temperature.

Statistical chicanery.

By Louis Hissink (not verified) on 26 Jun 2005 #permalink

Wrong Lewis. To understand why you have to read the paper http://pubs.giss.nasa.gov/docs/1999/1999_HansenRuedyG.pdf
. Briefly put the average over the period 1951-1980 is used as a reference. The period was selected because it was relatively stable temperature wise. This average is subtracted from the measurements to yield an anomaly. Anomalies for Norway and Australia can be directly compared.

By Eli Rabett (not verified) on 26 Jun 2005 #permalink