Hurricanes

Nate Silver over at the fantastic FiveThirtyEight has collected and analyzed data on whether (and how) the number of deaths or the damage caused by hurricanes correlates to how much media coverages the events get. He defined a "News Unit" as 10% of the entire media coverage for the time of the event (make sure read his original post for details).

Silver put all his data in tables, but I wanted to look at it. Hurricane Irene is shown as a larger dot in all the plots.

NUs vs Fatalities

The unfortunate outlier, not shown in this plot, is Hurricane Katrina, which had 1.56 NUs, but a fatality count of at least 1500. The color of the dots shows the reported maximum Saffir-Simpson scale of the hurricane, lightest being a tropical storm ("cat zero") to dark (cat 5).

NUs vs Damage

Again, the plot doesn't show Hurricane Katrina, which had an estimated damage of (normalized) ~95 billion dollars. It also doesn't show Hurricane Andrew, with an estimate damage of around 67 billion dollars. Same color scale as above.

NUs vs Category

The color scale here encodes the storm year, from (light) 1990 and earlier to (dark) 2011. From this very unscientific eye-balling, it seems that the Saffir-Simpson Scale is fairly predictive of hurricane news coverage.

NUs vs Year

Before anyone jumps to conclusions about the apparent abrupt change around 1998 (after that year, most storms appear to be widely reported in the media), let me clarify that I am almost certain that this dataset is not comprehensive. There are other storms which haven't been collected here and probably account for the effect.

Scatterplot Matrix

I apologize for the terrible tick labels; haven't found a good way to fix them. These plots show all the hurricanes, including Katrina. Notice how obvious an outlier Katrina is both in fatalities and damages. The next plot removes it.

Scatterplot Matrix without Katrina or Andrew

About

You can get the raw data by looking at the source. The plots were generated using Protovis. The scatterplot matrix, in particular, is essentially copied from Bostock's example. The year color scale of the category x NU plot is due to Brewer.

Back to my main page.