Can the crowd predict the weather?
Posted February 1, 2010
The computer models spoke. Then the Fish spoke. And we gave you a chance to speak, too.
More than 2,000 of you took up the challenge and made your forecasts of how much snow you thought we would see with last weekend's winter storm. That was in addition to the roughly 20,000 folks who voted in our very non-scientific poll on the front page.
Now that the snow and sleet have stopped and the totals are in, we have a winner! The National Weather Service recorded 5.0" of snow at the Raleigh/Durham International Airport from this winter storm. Four people — out of the total of 2,292 entrants — guessed the exact amount, so we drew one of those four at random to win the prize of a $50 gas card and a WRAL.com T-shirt. Congrats!
By the way, when the NWS tallies snowfall, accumulated sleet is counted as snow. They are different things, of course, and they accumulate at different rates. We often think of snow accumulating at 10" of snow for every 1" of liquid that you'd get if you melted that snow down, giving us a "snow ratio" of 10:1. Sleet, on the other hand, usually has a 4:1 ratio. (Just think how much snow we would have seen if things didn't switch over to sleet!) The bottom line is that the snow totals you've seen us show on TV and elsewhere include both the snow and the sleet that fell at those given locations.
On a more academic note, I find myself very curious at how the "crowdsourced" forecast turned out. There is some research suggesting that crowdsourcing may be somewhat useful in solving problems and answering questions such as these; although, exactly how useful and under what circumstances we can expect the best results are still to be determined.
I decided to see how our "crowdsourced forecast" turned out. Remember that, as of 6pm Friday, the official WRAL forecast had Wake county getting 6-9" of total accumulation. The National Weather Service's zone forecast for Wake county as of that same time was 4-8". (Remember, RDU Airport is in the northwestern end of the county, and that all of the guidance we had suggested higher totals north and west.)
To figure out more about what the crowd was saying, I put my spreadsheet to work. I plotted a histogram of the entries, counting the number of entries in each of 21 ranges, from "less than an inch" to "20 inches or more", and plotting them. The resulting plot shows a fairly nice bell curve, with a peak in the 7-8" bin (which I labeled as 7"). That lines up closely with the mean or average, as well as some other statistical values:
|Total number of entries||2,292|
|Fraction of entries within 1" of actual||424 of 2,292 (18.5%)|
A real, live statistician may have more to say about the results, but to my eyes, it says that the crowd was mostly optimistic about our snow chances, but not significantly more or less so than were we or the NWS. It's very likely the crowd's forecast was "seeded" with the numbers from WRAL, the NWS, and other sources, resulting in the similarity between the "official" forecasts and the crowd consensus. Still, in spite of that, nearly 20% of entrants in our contest were within ±1" of the actual snow total.
How did your snow forecast turn out? Do you want more or less snow next time around?