Dueling Forecasts

I once knew a person who was in charge of economic forecasting for a large corporation. I asked him about the models he used. He snorted with laughter and said that all he did was subscribe to lots of forecasts, take an average, and then write up a story consistent with the average.

I'm beginning to think we should do the same thing in our area with weather forecasts. Where I live has weather that is often difficult to forecast because of various lake effects and because a slight movement of the jet stream can throw off a forecast by a considerable amount. Fortunately, we have at least two regular weather forecast services for this area (and throughout Canada): Environment Canada, and The Weather Network. Here are their competing forecasts for this area for the next little while:

Environ. Can. ...........The Weather Network
Heavy snowfall ..........No explicit warnings
Snowsquall warning

over the next 24 hours:
25 - 40 cms snow ..........15 - 20 cms snow

Thursday night:
blowing snow..............variable cloudiness

The differences shown above seem typical. Often Environment Canada's forecast seem much more alarmist than those of The Weather Network. I've learned that most of the time The Weather Network is closer to what actually happens.

I can imagine that many of my readers will see this as yet another triumph of private providers over the bureaucrats. My own take is slightly, but only slightly, different:
I see The Weather Network as providing the equivalent of point estimates, whereas Environment Canada forecasts seem much more like "there's only a 20% chance the weather will get worse than this."

It's as if The Weather Network strives for being accurate, on average, whereas Environment Canada takes a more in loco parentis approach: we don't want to be responsible if you rely on our forecast and the actual weather is worse. This view is not unlike what Jack Quinn and I once referred to as "The Excedrin Theory of Bureaucracy" -- whatever you do, try to minimize the headaches to the bureaucrats.

Wouldn't it be nice if they'd give us probability distribution functions or confidence intervals or something with the forecasts? Yes, I know I'm dreaming if I think I would even be able to interpret such forecasts, but at least I am beginning to learn that using the two forecasts together gives me a better idea of the distribution function of a reasonable forecast.