post

Making all your points with graphs

James Hrynyshyn, a science journalist, pointed out on the Class M blog that a recent climate research graph was poorly designed.

The paper, Drought-Induced Reduction in Global Terrestrial Net Primary Production from 2000 Through 2009, demonstrated that anomalous CO2 and anomalous NPP (net primary production, described by Robert Simmon of NASA’s Earth Observatory as “a measure of the amount of carbon a plant takes from the atmosphere and uses to grow”) were negatively correlated. In other words, not only was increased CO2 not acting as “plant food,” it was undermining NPP overall.

In order to demonstrate just how strong this correlation was, the researchers inverted CO2 emissions on their graph:

Original NPP and inverted CO2 graph as shared on Class M science blog
Click for larger image.

As Hrynyshyn pointed out, this could readily lead to misunderstanding by non-scientists (or the occasional absent-minded scientist) to appear as though the CO2 anomaly was positively correlated to the NPP anomaly, instead of the opposite. That would be a significant misunderstanding, and in a politically controversial area such as climate change, a serious problem.

Robert Simmon, of NASA’s Earth Observatory, provided an alternative graph, clearly demonstrating the negative correlation:

Robert Simmon's version of NPP and inverted CO2 graph, as shared on NASA's Earth Observatory blog
Click for larger image.

But if I understand this correctly, the original graph had a useful purpose, it just went about it poorly. The point of the original graph was not to mislead about the positive/negative aspect of the correlation, but to demonstrate the strong level of correlation. This is a useful visualization in the right context, you just can’t do it by itself.

So why not add a third line? Something which showed the actual anomalous NPP and CO2 numbers with differently colored solid lines, as shown in the 2nd version, and added a clearly different third line (perhaps dotted, but in the same color as the CO2 to associate them), labeled CO2 (Inverted to demonstrate absolute correlation).

Revised NPP, CO2, and inverted CO2 graph, showing both actual data and degree of correlation
Click for larger image.

If you show the actual numbers clearly, then equally clearly distinguish the inversion, you can make both points without misleading, or allowing your graph to be misused.

Note: I commented this suggestion to the Earth Observatory post, and David Powell, another commentator, expressed concern the line could still be misunderstood. Powell wrote, “people would assume that a third line meant a third set of data and not just the same data plotted differently.” To show how I think it’s possible to avoid that, I created this example, which I think clearly distinguishes the inversion from the actual data.


Comments

  1. Alex:

    I’m not completely sold on this idea, but I like the approach. I think using a lightened and desaturated orange line would work better than a dashed line–the spaces between dashes create optical effects. I’d also label each line directly, instead of having a separate key.

  2. alex says:

    Thanks so much for responding, Robert. So you know, I did a dashed line because I pictured someone photocopying the graph in gray tones and losing the color distinctions. On consideration, though, that’s (a) probably an edge case and (b) if done, probably being done by someone who “gets” the graph. I like your approach.

    Labeling the lines within the graph makes excellent sense.

Speak Your Mind