How Come So Many Independent Papers Claim Hockey Sticks
Posted by Jeff Id on September 22, 2008
Ok, I have wanted to do this post since I figured out what is actually going on with the hockey stick graphs. The argument these days is that there are many papers which have repeatedly generated hockey sticks based on temperature proxies. First let me explain what a temperature proxy is and why these papers claim that we are warmer than the past 2000 years. If you already know this stuff skip to the third paragraph.
A temperature proxy is a measurement of something which is belived may be related to temperature but no-one is sure. Things like tree ring widths and isotope ratios in various objects from stalactites to mussel shells and ice cores are used. Scientists compare the graphs of these various items to temperature graphs and do statistics to see if they actually might be temperature. I personally believe they have little or no temperature signal in the graphs whatsoever but pro-global warming scientists use difficult to understand statistical techniques to sort and throw away data which don’t correspond to temperature averaging the remaining data. Much of the data is thrown away during this process. The latest hockey stick threw out 60% of its data.
With that in mind, I took the time to make up an experiment. First, I made what is called a red noise generator. It is a technical term meaning random data which demonstrates trends over short term periods.
Below are 20 series of red noise generated by my computer, I highlighted a few lines so you could follow them.
This data is entirely random. If you average enough random data you should get a flat line. So using the power of today’s PC, I made 10,000 curves like the ones above and averaged them below.
It is a pretty flat line as you can see.
Now red noise is pretty analogous to the natural noise in tree rings and other temperature proxies. For instance, a tree grows nearby to a measured tree blocking light on one side. At first there will be little problem, but as the neighbor tree grows its shadow casts over the first tree creating some ‘stress’ and slowing its growth. Another example might be drought conditions for a period of 20 years. These are just examples which create shifts in the tree ring widths over time, since we are looking for temperature signals these effects are red noise. The tree most likely also changes growth rate based on temperature, and thus the study of dendroclimatology was popularized.
The problem is, how much does temperature affect tree growth. Does a little temperature rise make the tree grow faster and a big temperature rise make the tree grow slower. It seems reasonable but very little experimental work has been done on this. The only real methods which I have read have to do with comparing the graphed shape of tree ring widths to the shape of measured temperature. To say this comparison is full of problems is an overwhelming understatement.
Let’s play with some red noise. I took the very same series as above and added a fake “temperature signal” to it. This is the signal we will attempt to extract from the random data. The signal is shown below. Temperatures are flat except for between 1200-1300AD.
I added it to all 10000 series above.
The random + temperature signals look like this.
If you look close you can see an up shift in the series from the first graph to the above graph from 1200-1300AD.
I averaged all 10,000 series in the next graph including the temp data. Computers are pretty awesome, imagine doing this 30 years ago!
As most of us would expect the temperature signal has the same magnitude as the ‘signal’ data, everything else is nearly flat. I took a random 2000 series from the temperature curve above to show that any random sampling of temperature will produce a similar shape. —- It does.
Sure enough, it has the same shape and amplitude as the curve above. You can see additional noise from a less complete dataset. (i.e. the more curves you average the smoother it gets)
What paleoclimatology does though is to look for temperature trends in the data. Clearly there is no trend in the last 100 years in the data above, yet what happens if we look for a trend. This is where paleoclimatology goes wrong. THIS IS THE ONLY SCIENCE FIELD THAT I KNOW OF WHICH DOES THIS.
If we sort the data above, the same random meaningless data according to the top 5% of the maximum upward slopes over the last 100 years. For those who know math, I fit a linear least squares curve to the last 100 years and took the top 500 maximum slopes. And for those who live in statistics and understand the limitations of this process, I am telling you that this is not significantly different from the EIV comparison or the non-centered (I like that) PCA which Mann used in the 98 paper. Sure the result is not the exact same, but looking for a trend in a near random set has huge pitfalls.
Anyway the graph is here.
Now for the rest of us who don’t want to learn every detail of climatology and how every friggin paper makes a hockey stick this should be a meaningful graph. I just took random meaningless data and found a huge very steep up slope in temperature in the most recent times! This is exactly what Steve McIntyre showed to the world in his papers and which he referrs to as mining for hockey sticks.
He presented the best evidence which could be compiled to the paleoclimatologists who rely on this method that it was faulty and not surprisingly they didn’t change their ways. If you look past the red and grey lines in the graph below which are the only measured temp curves plotted on the latest hockey stick, you can see that the other data doesn’t make a spike in recent times. It looks a lot like my graph above with a bit more red noise.
You have to look beneath the temp data (red and grey lines), but the resulting trend dips lower than the red series because of the comparison. This is similar to the above graph dipping below zero for the 1900 time frame data.
I’m not sure if I can finish the math this week because I am working on other things, but i will show later how p value correlations will create the same effect. Correlation is the rational that paleoclimatology uses for its excuse to discard otherwise valid data. From my experience processing data I can say for certain, p value correlation can find a hockey stick in this random data also.
This isn’t the only reason that “independent papers” get hockey sticks, It also happens because they tend to use exactly the same data sets. I had to add that because it is a big point by Steve McIntyre at Climate Audit!
You then get a bunch of Peers reviewing papers which sort data with similar processes to what they did in their own last paper.
I need to thank everyone for your support in reading this article. This presents the foundation for my second post on the subject which delves deeper into the math.
It’s slightly more complicated but it reveals a more serious problem in the creation of hockey sticks. The demagnification of historic data in favor of recent trends. The effect reaches across every temperature construction which employs data sorting and calibration to locate the hidden signal.