Blog Archive

Monday, March 31, 2014

ROFL - Unnatural disaster: Roger Pielke, Jr.’s writing is an embarrassment to Nate Silver’s new venture

by David Auerbach, Slate, March 31, 2014

Nate Silver at SXSWi in 2009.
Nate Silver is still backing the wrong horse.
Photo courtesy Randy Stewart/Flickr
I come to praise Nate Silver, not to bury him. In this case, I want to congratulate him for taking the first step toward the eventual firing of Roger Pielke, Jr., author of one of the most heavily criticized pieces to date on Silver’s somewhat beleaguered new site, FiveThirtyEight. Pielke’s March 19, 2014, piece, “Disasters Cost More than Ever—But Not Because of Climate Change,” has been slammed for faulty data, ideological bias, and poor statistical work, leading Silver to publish a conditional defense of the article. Silver admits to some sloppiness in the piece and has said that FiveThirtyEight will publish a rebuttal. These are good steps, but Silver is still backing the wrong horse, and the sooner he dumps Pielke (who’s also a professor of environmental studies at the University of Colorado, Boulder), the better.

Issues around climate change and disaster statistics are so complicated—far more complicated than election polling—that you might be tempted to think that, as Silver says, “The debate is hard for us to adjudicate without turning to experts for help.” Silver is wrong. On Pielke’s own terms, even under the most charitable interpretation, his article is an embarrassment. It does not take a trained statistician to see what’s wrong with Pielke’s article, just some old-fashioned critical thinking.
Let’s assume, for the sake of argument, that Pielke’s stats on natural disaster frequency and natural disaster cost are absolutely right. (Granted, no one seems to think this.) Let’s only focus on Silver’s statement: “The central thesis of the piece was that although these costs are increasing, the rise can be accounted for by the growing wealth of the global population, rather than by a rise in the number of disaster events due to climate change.” The article fails to make a coherent case even for that modest thesis.
Here is Pielke’s “killer” graph, intended to show that disaster costs are not growing relative to GDP:
140331_$BOX_global-disaster-losses
This graph only spikes when bad things happen to white people. (Or Japan.)
Consider what is being measured: the aggregate total of disaster costs vs. the global GDP. As we all know, global wealth isn’t distributed evenly. “Wealthy countries hold all of the sway in worldwide cost estimates,” Pielke writes. The 2004 Indian Ocean earthquake and the 2010 Haiti earthquake, the two deadliest disasters by far during the time period of the graph—both in the hundreds of thousands of deaths—don’t produce any spikes whatsoever, and belong to “average” years. Hurricane Katrina, which resulted in around 2,000 deaths, produced that large 2005 spike. This graph isn’t measuring deaths, of course; just economic damage to property and productivity.
So let’s say Lex Luthor causes California to fall into the ocean. The graph will spike like mad. Let’s say Dr. Magneto causes half of Africa to fall into the ocean. The graph will show a mild bump. Pielke decided that aggregating all the costs together for world GDP vs. world disaster costs would be more illuminating than breaking it down even on a continent-by-continent basis. I won’t speculate on his motives here. (Everyone else has already done that for me: Climate Science Watch concludes his goal is to “dismiss the wide array of evidence linking climate change to extreme weather,” while climate scientist Dana Nuccitelli says Pielke is known for “downplaying the links between global warming and extreme weather.”) But this graph hardly seems like the most relevant measure of the impact of natural disasters.
In an attempt to make this graph meaningful (or at least less offensive), Pielke pulls a bait and switch. He writes that increased disaster costs mean that fewer people are dying: “The data show an inverse relationship between lives lost and property damage.” Pielke seems to be implying that as GDP goes up, property damage goes up—even assuming that the number of disasters stays constant—but the lives lost in disasters will go down. There is a correlation, Pielke says, between being rich and being safe in a disaster; the richer the country, the fewer the lives lost when a disaster hits.
“There is some good news to be found in the ever-mounting toll of disaster losses,” Pielke writes; richer countries are “better able to deal with disasters”—Pielke doesn’t say why, but presumably through better infrastructure, safer buildings, greater population mobility, early warning systems, and the like.
So what started as an argument that increased disaster costs aren’t as bad as they seem has turned into an argument that increased disaster costs are actively good. We should be happy if disasters are destroying more Fabergé eggs, because people with Fabergé eggs probably live in earthquake-proof houses.
So Pielke is arguing for two opposing theses: First, that disaster costs haven’t actually gone up that much (it’s not climate change’s fault!); and second, that disaster costs are still going up but that’s a good thing (fewer lives lost!). In other words, it’s a Goldilocks situation: Disaster costs are going up just right.
This is preposterous. The graph only shows two variables, while Pielke is making an argument about three variables: disaster cost, economic growth, and disaster occurrence. The graph conflates disaster cost with disaster occurrence, making it impossible to individuate trends in either. A spike shows that an expensive disaster happened (bad!) but that disaster costs went up (good!). No spike shows that only cheap disasters happened (also bad!), but disaster costs stayed down (also good!).
Compared to the Munich Re data or the IPCC report on which Pielke drew, Pielke’s graphs look closer to USA Today infographics than serious statistical work.
While Pielke is fiddling with GDP numbers and celebrating the wealthy, the second half of the IPCC report, just released yesterday, concludes that climate change risks destabilizing human society and that the projections are grave. Evidently that part didn’t grab Pielke.
140331_$BOX_temperature-anomaly
From the IPCC 2013 report
140331_$BOX_too-complicated
Too complicated for FiveThirtyEight?
From Munich Re
At first glance, Pielke just seems to be dumbing down the material, but the internal contradictions in his article make it look far closer to intellectual dishonesty, collapsing variables and muddying the waters. (For example, why is he using a graph whose two biggest spikes come from earthquakes when he’s ostensibly talking about climate change? Is he proposing a heretofore unknown link between earthquakes and climate? Scoop!) The good thing about intellectual dishonesty is that errors beget errors. Sneaky statistical manipulation often bubbles up into conceptual incoherency, as it did in Pielke’s article and now in Silver’s defense of it. There is no need to debate the evidence because the article collapses on its own terms.
Silver seems to realize that he fell down here, and so I hesitate to criticize him too heavily. Silver constitutes such an improvement on horserace journalism that you can bet Politico and Daily Caller are rubbing their hands with glee to see Paul Krugman and Nate Silver turning on each other—it’s even better than Twitter turning on Stephen Colbert. Statistics experts Cathy O’Neil and Deborah Mayo have made some serious criticisms of Silver’s own work, but given the state of much punditry today, the perfect remains the enemy of the good—or at least the enemy of the better-than-Politico. When Leon Wieseltier praises George Will (a pundit whose views on climate change make Roger Pielke look like Al Gore) as superior to Silver, it’s enough to make me drop to my knees and beg Silver not to throw in the towel. Please, Nate, cut Pielke loose and stay the course.
http://www.slate.com/articles/business/moneybox/2014/03/nate_silver_climate_change_denial_it_s_time_to_dump_fivethirtyeight_s_roger.html

No comments: