The only thing we can be sure of about the future is that it will be absolutely fantastic. So if what I say now seems to you to be very reasonable, then I have failed completely. Only if what I tell you appears absolutely unbelievable, have we any chance of visualizing the future as it really will happen.

- Arthur C. Clarke (h/t Brin)

Tuesday, March 31, 2009

And What's Wrong with This Picture?

I accidentally hit a site that is promulgating this graph; not the first time I've seen it.
Note that it's perfectly true, and that the vertical scales and start and end points have been carefully chosen to yield a misimpression. Does this constitute lying? To a political or legal mind, I think it doesn't.

(Update: also note that the two curves have different time filters applied. Consider why this would be so. Hint: it is related to the choice of vertical scale.)

In a blog comment exchange here, I casually mentioned "cherry picking" and "Ted" elaborated:
The summary reference to “cherry-picking” says a lot. What it illustrates, more than anything, I think, is the different sorts of argument that count as acceptable in science and in political discourse.

In science, the assumption is that it’s not okay to treat evidence selectively. You’re supposed to *try* to account for all the evidence.

In political discourse, I’m afraid, the de facto assumption seems to be that it’s fair to pick up whatever data point happens to be handy and throw it at your opponent, while (of course) ignoring and evading the data points they throw at you.

George Will’s recent column was an excellent instance of what happens when you take the norms of political discourse and apply them to science. Will may have thought he was just “spinning” — which is more or less what he gets paid for — but he was spinning a topic where “spin” counts as culpable distortion.
Update: Here's the same data without the three bugs. Consider plotting the above graph for 1996- 2006 instead of 1998-2008.


Stop the presses, huh?

See also How to Tell Different Stories with the Same Data.

12 comments:

gmcrews said...

Hi Michael,

You say the above chart does not "constitute lying" to "a political or legal mind". Does that mean you think it is lying to a scientific mind?

How can this be? As "Ted" elaborated: "In science, ... [y]ou're supposed to *try* and account for all the evidence."

So I did. I had a Bayesian prior that increasing CO2 levels mean higher temperatures, maybe nonlinearly higher. I would not have expected the above chart. I could make up some numbers, but obviously after looking at the chart above, my conditioned belief is a bit lower. Isn't yours? Only a prior of 0 or 1 would be immune to this new evidence.

The scientific method is a process that if politicized, turns into something other than the scientific method. We will go wrong if the result we want becomes more important than the process we use to get .

So I think all this worry about "misimpressions" is getting us off-track. We should thank those who try and remind us we could be wrong. After all, the only rationally way convince the public we are right, is for them to see us rationally adjust our beliefs when we are given unexpected data. The above chart should rightly (Bayesian) make the public a bit more skeptical of the relationship of CO2 and temperature. And we are too. But just look at this other evidence...

Michael Tobis said...

It deliberately presents a misimpression for two reasons

1) the slope of the CO2 curve is arbitrarily stretched; to match the proposed sensitivities it should be very nearly flat.

2) The start date and end date are very carefully chosen to make it look like a cooling.

Indeed, if this were all the data, it would affect our view substantially. But because it isn't, because it is carefully chosen in advance for polemical reasons, it should not be considered as informative.

Strictly speaking you might contrive a very complicated Bayesian statistic in which this had a very tiny impact on your posterior estimate but it would be a silly contrivance, perhaps useful as an exercise but not meriting attention at outreach websites. If you did that correctly, it is not obvious to me whether that would raise or lower your expected sensitivity.

The question would be something like "choose a peak and a trough from the record spanning more than five and less than twenty years ending within the last five years so as to minimize the slope of the linear fit. How does that slope affect your estimate of sensitivity"?

In any case I can see nothing to justify the choice of vertical scale on the graph.

You seem to be arguing for argument's sake. Please be serious.

Michael Tobis said...

Note also that you have monthly temperatures and a CO2 curve that at best has the annual cycle subtracted.

What would the graph look like if the time filters applied to both curves were comparable?

Scruffy Dan said...

Robert Grumbine has a post on an appropriate way to determine if CO2 is correlated with temperature. It is well worth a read.
http://moregrumbinescience.blogspot.com/2009/03/does-co2-correlate-with-temperature.html

gmcrews said...

Hi Michael,

(UPDATE: to my original comment, after your UPDATE to your original post.)

I can be serious and still insist on not insulting anybody. I think it important because consider how you would respond if I commented on your second graph (the one labeled "without the three bugs") in the following way:

"Note that it's perfectly true, and that the vertical scales and start and end points have been carefully chosen to yield a misimpression. Does this constitute lying?" "Indeed, if this were all the data, it would affect our view substantially. But because it isn't, because it is carefully chosen in advance for polemical reasons, it should not be considered as informative."

I am certain such a response would harden most third-party viewpoints or worse. But you're a smart guy and disagree. So I won't bug you about it anymore. Good luck in your approach. We need it.

Steve Bloom said...

gm, if our starting point was the view that sensitivity is so large or natural variability is so small that temps can be expected to increase over pretty much any selected short period of time, which is basically what your prior means, then I would completely agree with you. We didn't start out with that view, though, although it's hard to argue that many uninformed people behave as if they had. Actually probably most such people have a "prior" more along the lines of "it was cold this winter where I was, and it seemed to me to be unusually cold, so global warming must be over."

It will be interesting to see the responses when the next record warm year happens.

Steve Bloom said...

I'm not certain, but it looks as if "climatologist" (sigh) Joe D'Aleo may have been looking at a version of this same graph in the Media Matters clip featured in the next thread.

Marion Delgado said...

Science and politics really don't mix.

Marion Delgado said...

I think we need to revoke certain people's big-words privileges.

Just to tide them over, I'll add that I mean the denizens of the subpontic denialosphere.

David B. Benson said...

Scruffy Dan --- Your link is broken. You have to use the HTML "a" tags.

Scruffy Dan said...

Ya I know I just got lazy. the link does work if you copy and paste it into your address bar. But just in case others are lazy like me here is a proper link.

Joel said...

Steve:

Actually, as near as I can tell, Joe D'Aleo (and/or his ICECAP website) is the originator of this and similar graphs, or at least a major promulgator of them. There seem to be lots of different versions with different start dates, slightly different end dates, different temperature sources, linear trend lines plotted or not plotted, etc. But they all have that same color shading to them and they all align the CO2 and temperature scales for a transient climate response of ~9 C per CO2 doubling!

This is the all-time favorite plot of one "Smokey" over at Anthony Watts's blog and no matter how many times I try to explain the deception to him, it never makes a bit of difference.