Tuesday, March 10, 2009

Misleading yourself with graphs

I've mentioned before my appreciation of the book How to Lie with Statistics by Darrel Huff. One of the major sections regards graphs. Part of doing science is honesty, and, as Richard Feynman noted, the easiest person to fool is yourself.

Friday, I was doing some browsing and encountered co2sceptics. The graphs on that main page (at least on March 6th, 2009) were plots of the Hadley temperature record, from 1978 to 2008, and the atmospheric CO2 concentrations as measured at Mauna Loa from 1958 to 2008. That struck me as odd. Hadley goes back much farther than 1978, to 1850 in fact -- and co2sceptics link to this longer record so they must know that. Why show only 30 years data when you have almost 160? If nothing else, one of the How to Lie with Statistics methods is to show different time lines when comparing different information sources, and here we've got 30 years being shown against 50 years.

One thing to do, then, to try to avoid misleading yourself, is to plot 50 years for both. Here are the two plotted over the same time interval:

Let's see, the CO2 is rising steadily, and there's weather happening on top of temperature rising. Rather different from their conclusion (the quote is below).

But let's not make the same error of using too little data. Plus, with only 50 years of data, and temperatures having a climate period of 20-30 years, we might only be looking at a minor anomaly. Hadley goes back to 1850. Can we find CO2 information back that far? Yes, using the CO2 trapped in ice core bubbles, say at Siple. Here are the plots for 1850 to most recent, where I use Siple for CO2 until the start of Mauna Loa:

Now here's the text they introduce their graphs with the different time spans with:

If you look at the Global Temperature Anomaly graph below you will notice that temperatures peaked in 1998, and have been in general decline ever since, unlike the temperature figures you will see that CO2 output is still increasing.

Now ask yourself a simple question. Is CO2 driving the world's temperature changes?

It would appear not!

Emphasis theirs.

Actually, given the data types and sources they chose and method of comparison they chose, it looks like CO2 does indeed drive climate on longer periods than individual years. All we had to do was show data for temperature over the full period of the CO2 record they showed. Extending the CO2 record with ice cores only makes that connection look stronger.

Now, since I think you need 20-30 years of global mean temperature data to be talking about climate rather than weather, I wouldn't trust that method of analysis very far at all, whether by eyeball or doing a regression.

Since I'm an optimist, I've sent a note to them about this post. If it's just a matter of a mistake, well, honest people can make mistakes. They can then correct their figures and update their comments to reflect a significant role for CO2 according to their already-shown methods of analysis. And if they're not concerned with honest discussion of science, they wouldn't be the first.

Mauna Loa CO2 data
Siple CO2 data
Hadley Temperature data


Scruffy Dan said...

This reminds me of a recent silly graph put out my Monckton, where he overlaid temperature and CO2, while scaling the y-axis to show the temperature trend line as being almost flat (though still increasing) and the CO2 trend line increasing at a much greater rate. Thus 'proving' that the two aren't correlated.

Of course given the source, I am not optimistic at all that this was just an honest mistake. Especially given his liberal interpretation of other graphs.

Oh, please let us know when/if you get a response from co2sceptics.

Penguindreams said...

I've invited co2sceptics to comment here directly. Though if they prefer an email to me, that's fine.

The Monckton graph you describe sounds like one that icecap.us was using, and which I had some comments on.

Philip H. said...

What's funny/ironic/sad about all thsi is that the in the temperature graph you showed, it's clear to anyone who has had even a single semester of statistics that 1998 an outlier, especially if you fit a trend line to the graph. it never ceases to amaze me how these folks will do that, and react with HORROR when their error is pointed out to them.

Scruffy Dan said...

Ya, Monckton's graph is similar to the one from ICECAP, but even less convincing since it shows both the temperature and CO2 trends as increasing.

Either way neither of these graphs is proof of anything other than dishonesty from Monckton and ICECAP. And depending of their response from CO2sceptics as well.

quasarpulse said...

The other problem is their use of the dishonest "connect the dots" line on the Hadley data instead of a proper trendline. As I believe you've mentioned before, there's no way to talk about a meaningful climate trend on a scale less than ~20(?) years, which makes the "decrease" since 1998 meaningless.

Hank Roberts said...

More of use here (I've pointed to these before) from Stoat and Atmos on five year trends. The search also finds some muddlefuddle from WTF.

Penguindreams said...

Gary, your post is not appropriate to this thread -- belongs over in the CO2 correlation. Please resend to that thread, or send a note that it's ok with you for me to copy it over there.

Hank Roberts said...

Another trend-chart-interpreter:
"Note the sudden directional changes that occur in 1946, 1976 and the early part of the last decade. These are not slow and gradual shifts, they are sudden alterations in directions."