Statisticians Cool Down the Climate Controversy
An Associated Press investigative report debunks claims from global warming skeptics.
Lynne Peeples • October 28, 2009
Why didn’t I think of this? I know an estimated 90 percent of scientists agree that our earth is warming. I’ve heard the conservative claims of just the opposite: that the climate is cooling. (They point to what they see as a downward trend in temperatures over the last decade.) I’ve also seen the results of the latest Pew poll suggesting that only 57 percent of Americans believe in global warming—a drop of 14 percent since April.
I don’t need my math and statistics degrees to recognize that these numbers all add up to dangerous misinformation. But whenever I’ve tried to convince a skeptic that, yes, the climate really is warming—not cooling—the conversation usually dead-ends. Both of us invariably accuse the other of approaching the data with preconceived notions, cherry-picking our numbers.
The Associated Press’s ingenious solution: Give the dataset to some independent statisticians to analyze, but take out the temperature and year labels—thereby also eliminating any emotional prejudice.
In the blind stat test unveiled this week, four out of four statisticians chose global warming over climate cooling.
How could any previous analyses, regardless of their emotional backing, have found a cooling trend? It’s all about the data you choose, and how you slice it, explains the AP. There are many opportunities for intentional or unintentional manipulation, including the choice between ground and satellite data (the former tend to be warmer; both were given to the statisticians), the range of years, and how those temperatures are averaged.
As the AP story explains, the preferred dataset for many skeptics begins in 1998—a scorching hot year—and includes only satellite-recorded temperatures over a 10-year period. In this relatively small dataset, the influence of that single year can be huge. It can be enough to produce a statistically weak downward trend through the highly variable temperatures. But start the analysis in 1997 or 1999, and the 10-year trend tilts up.
A fairer assessment would be to analyze what are called 10-year “moving averages”. In other words, average temperatures between 1997 and 2006, 1998 and 2007, and so on. Then look for trends in those averages. This way, any one anomaly has less sway. Such an analysis, according to the AP, shows temperatures over the last five years were higher than in any previous years.
Of course, even this is too narrow a view. Climate trends over Earth’s history, beyond just 10 years ago, offer a far more telling tale. “To talk about global cooling at the end of the hottest decade the planet has experienced in many thousands of years is ridiculous,” Ken Caldeira, a climate scientist at the Carnegie Institution at Stanford, told the AP.
So why does the talk continue? As more and more scientific data pours in, all pointing to rising temperatures in the decades ahead, why are fewer and fewer Americans accepting it? What will change their minds?
Some suggest it will take more extreme events, or other health concerns that affect people personally. In fact, a series of five research papers will be published in the November 27th issue of the journal The Lancet, nicely timed just a couple weeks before talks begin in Copenhagen. The studies will lay out various health effects of climate change—from increases in the number of heat waves and stagnant air masses to a potential rise in insect and water-borne diseases. The issue will highlight how reducing emissions could have a positive impact on public health.
I’m still (irrationally) hopeful something more “innocent” could do the trick. Maybe this new video on Huffington Post, for example, will capture the attention of some of the more visual skeptics: “Supermodels Strip for Climate Change.”
Of course, this may not necessarily motivate viewers to fight warming, but at least it might make them believers.
Folks can download that data and plot it using Excel. That’s what I do because the climate issue is so politicized that there is often fudging. Take a look at the chart above. The blue plot is a moving average. Under that are red bars representing the separate years. Notice that each of the last four years is cooler that the year before it. That is a short term cooling trend. What it means is subject to debate. The fact that it is there, is not. Now look that the blue line. It isn’t reflecting that cooling trend because the averaging span is longer than the trend. That is deceptive. And I believe that it is intended to be deceptive. The AP article is fudging in a very similar way. That’s clever I suppose but should deception be a part of science?
re: #1: This article actually addresses your criticism very directly. Yes, it is true that displaying a 10 year moving average is somewhat subjective. The point, however, is that a multi-year averaging is exactly what is needed to detect trends in climate, which by definition express themselves over multiple years.
A 4 year moving average plotted for this data would still show the same long term trend. The difference would be that the blue line would be less smooth and tip down more at the right side of the chart. The overall trend would remain. Looking at the annual mean data there are plenty of segments where someone could say “look at the downward trend here” (for example, the 1940s data). The issue, metaphorically speaking, is whether such trends from peaks to troughs are reflective of the actual landscape or the anthills along the way. It is not a deception to state upfront that a moving average was chosen based on known variabilities and then analyze the data with the averaged data. Why would you choose to define a climate trend based exclusively on four years of data, given what is known about inter-annual variations in weather (such as El Nino/La Nina patterns), while ignoring 120 previous years of data?
Regarding R. Jackson’s comment at #2.
A 10-year centered moving average cannot show a either a warming or a cooling trend for the last 5 years because a centered average must average the ten years using the data 5 years before and 5 years after each point. A 4-year centered moving average would have to be stopped 2 years short of present.
A linear regression can be calculated for the previous 5 years. With most data sources, a linear regression starting at some point before 2004 but after the strong El Nino in 1998 and the strong La Nina that followed it, will have a cooling trend. That is apparent in the red data bars above but not in the running average. Obviously a 4 or 10 year leading average (such as is the standard with Excel) or a trailing average would poorly represent the data because of the displacement.
If we are going to point out that it has been cooling for 4 years (actually longer than that), then the question would have to be whether there has been a cooling trend for the last 4 years. If one does a linear regression starting in 1900 (for example) it would show the warming trend. It could cool for more than a decade (probably several decades) and the regression would still show a warming trend. If we did, would it still be fair to said that it is warming even when it has been cooling?
It is reasonable to question whether a short-term cooling trend is significant. But I don’t think that we should pretend that the short term trend isn’t there because it isn’t a long-term trend. Every long-term trend started out as a short-term trend. We cannot know which is which except in retrospect.
In the caption under the chart, they say that “Experts agree that the Earth’s climate has been steadily climbing since 1880.” A linear regression across that span would show warming. But look at the chart. From around 1910 until around 1940, it warmed at a rate and magnitude comparable to the period from around 1975 to 2004. But the increase in atmospheric CO2 was only around one third of the increase more recently. The IPCC claims that “most” of the warming of the last 30 years or so was caused by increasing CO2. But look at the Pacific Decadal Oscillation (PDO) and the Atlantic Decadal Oscillation (AMO). There is a much better correlation than with CO2 over the last 120 years.
I don’t advocate ignoring the last 120 years of data. I just don’t think that we should ignore the last 5 years of data either.