Relying on accurate weather forecasts & Fumbling the Statistics

Tim Harford has an interesting blog on weather and the accuracy of forecasts. However, he does not look at two aspects.

1. The Consequence of Reliance

Harford comments

“A recent case in point: Bournemouth’s woes during the bank holiday at the end of May. The Met Office predicted storms, but the beach resort in fact enjoyed the sunniest day of the year. Bournemouth’s tourist office reckons the town missed out on at least 25,000 visitors and more than £1m of revenue as a result. Subtler losses and gains were registered by the would-be tourists, and the lucky ones who enjoyed both a sunny day and a quieter beach.”

Imagine if this had been the USA. Would the BBC have received class-action suits unless every 2 minute weather forecast had the sub-titled with long sentances about only being an opinion, and no responsibility held for the accuracy. Would they also become increasingly like prophesies from the Delphic Oracle?

2. The Statistical Interpretation.

Harford says

MIT economist Michael Greenstone has studied the impact of local temperature surges on deaths in both India and the US. He calculates that a year with one extra “heatwave” day – temperatures above 32°C instead of 12°C-15°C – would raise the annual death rate by eight per million in the US. In India, the temperature vulnerability is more than five times higher, notably in rural areas where agriculture suffers and wages drop.

The population of the USA is around 320 million. So an extra day of heatwave will cause 2.5% of the population to drop dead.

In India the 5 times higher vulnerability means that one-eighth of the population could be wiped out. Or is could mean that the death rate is only one third higher, allowing for the population is 4 nearly times that of the US.

Maybe the USA figure should be 8,000 and not 8,000,000. It is a shame someone cannot sense check these figures, as Harford has some interesting insights on his blog at the FT.

Previous Post
Comments are closed.
%d bloggers like this: