Mathematical Malpractice Watch: Weather Fatalities

This came to my attention a month ago. I drafted a post, forgot about it in the election/migraine event horizon but now want to get it out my drafts section. I think it’s worth posting because we are likely to hear more of this from the more hysterical environmental wing.

The chart, from Ezra Klein’s usually excellent Wonkblog, purports to show a steep rise in weather-related fatalities in recent years.

It doesn’t show anything of the kind.

First of all, what it shows is a slight decline or flat trend with a few recent spikes caused by a 90’s heat wave, Hurricane Katrina and last year’s tornados. Now maybe you can argue that we should pay more attention to these in the era of global warming because they may be related (or may not). I agree. However, the long term trend in almost all categories is down — way down. Deaths from lightning strikes are down by over two-thirds over the last 70 years. That’s real progress.

But the progress is even better than the graph shows. The graph makes a huge blindingly obvious error; one that Klein’s readers jumped on immediately: it does not account for population growth. The first data point is from a sample of 140 million people while the last if from a sample of 310 million. To compare raw figures is simply ridiculous (and, indeed, Klein’s co-blogger later tweeted a version with death rates that was far less dire and showed dramatic declines in weather-related fatalities).

The third problem is less obvious but potentially the worst one. The plot includes deaths from heat, cold, “winter fatalities”, rip currents and wind. Heat deaths are particularly important to the point Wonkblog is making since, presumably, global warming will result in more deaths from heat waves and drought.

The problem is that the NOAA, from whose data the graph is taken, did not track heat deaths until 1986. The same goes for many deaths in the “other” category. Cold fatalities were not tracked until 1988. Winter fatalities until 1986. Rip currents until 2002. Wind deaths until 1995. No correction, none whatsover, is made for the incomplete data that spans the first five or six decades of NOAA’s sample.

It is simply not sensible to treat the data as though there were zero deaths from heat and other categories before the mid-1980’s. In fact, there are many reasons — the spread of air-conditioning for example — to suspect that heat-related deaths were much much higher in the past. It would defy common sense for the sharp reductions in fatalities from tornados, hurricanes and lightning (not to mention earthquakes) to not reflected in the statistics for other weather-related deaths.

But let’s not assume. Let’s go to the record. The data start in 1940, which usefully omits one of the greatest environmental calamities in American history: the Dust Bowl. Thousands died; at least 5000 in one 1936 heat wave alone. Another massive drought hit in the 1950’s. A 1972 heat wave killed 900 people. A 1980 heat wave killed 1700 people. All of those happened before the NOAA tracked the number of heat-related deaths. None are in the sample.

To be completely honest, the NOAA data seems a poor resource for this kind of study. It apparently does not include the 1988 drought, recording only 47 heat-related deaths in that two-year period. But it does include the 1995 and 1999 heat waves. I have no idea what their criteria are. I suspect they are counting deaths from specific short-term heat waves rather than broad massive events like the 88-89 drought. That’s fine as far as it goes. But if your attempt to quantify long-term trends in weather-related deaths ignores droughts; if it ignores the God-damned Dust Bowl, I would submit that you are looking at the wrong data.

So, in the end, the claim that we are getting more weather-related fatalities than ever is, at least in this case, based on a heavily biased poorly understood sample that barely supports the conclusion