Eric Laermans: The Launch of fivethirtyeight.com and Climate Change Disaster Weblogging: (Trying to Be) The Honest Broker for the Week of March 29, 2014: "I can't help but wonder why nobody is pointing out...
...that including costs incurred from natural disasters completely unrelated to climate change (e.g. earthquakes or volcano eruptions) also is a deadly sin against correct data-driven analysis. These non-climate-related disasters add a lot of noise to the data and make it needlessly harder to identify the impact of climate change on the global cost due to natural disasters, unless of course it was the author's intent to muddle the analysis.
Touché...
In fact, when you look at the data Pielke uses to establish his claim that costs of natural disasters are (a) growing over time but (b) not growing faster than one would expect given growing wealth, you wonder why Nate Silver or somebody else didn't bounce the article immediately.
The big enchilada in Pielke's 1990-2013 graph is the 2011 Honshu Earthquake:
Conditional on the Honshu Earthquake happening in his sample, it is just pure chance that the Honshu Earthquake happened at the end rather than at the beginning. And if the Honshu Earthquake had happened in, say, 1991, it would have done more damage: the Japanese economy has not grown materially since 1990, and Japanese earthquake safety standards have materially improved.
When the shape of one's quantitative argument about the costs of global warming hinges on a huge and damaging earthquake occurring at the end rather than near the beginning of one's sample, one can be said to be doing many things--but not data science...