Monday, March 03, 2008

February Corbynwatch

Feb was supposed to be warm and wet. As I said earlier, a CET temperature in the range of 4.5-5.5C would fit the bill - and this time it seems like he got it right! This page actually gives a value of 5.51C but it seems to be biased a little high, so I'm happy to count that as a win for him. The rain forecast didn't turn out quote so well though:

CorbynRain: 160%Sun 60%
RealityRain: 60%Sun: 160%


(% of normal values)

But I'll only count that as a single failure since rain and sun are obviously negatively correlated (rain and temp is not so clear: Jan was warm and wet, Feb was warm and dry).

Looking just at these monthly means for Jan and Feb (temp and rain), Piers scores 1 out of 4 - Feb temp is the only one he got right. His claimed probabilities for these forecasts were between 75% and 85%. Using a typical value of 80% for simplicity, the probability of different outcomes can be calculated as:

Win-Loss4-03-12-21-30-4
Probability41%41%15%2.5%0.2%


(yes, I've learnt how to put tables in to Blogger without messing up the spacing).

So either he's been extremely unlucky...or his claimed probabilities are wrong. The lesson here is that at a predicted 80% probability, even with only a modest sample you have to get a fair number majority right, not just the occasional lucky hit. With more months, it will get increasingly implausible that his failures can be put down to bad luck. I haven't even bothered with the string of storm forecasts that have failed so far.

March's forecast hasn't appeared yet.

8 comments:

skanky said...

"But I'll only count that as a single failure since rain and sun are obviously negatively correlated (rain and temp is not so clear: Jan was warm and wet, Feb was warm and dry)."

Interestingly (or not), the winter as a whole, was above average in temp., precipitation and sunshine.

See:
http://www.metoffice.gov.uk/corporate/pressoffice/2008/pr20080228.html

Hank Roberts said...

Is Corbyn the same as weatheraction.com?

Hank Roberts said...

Is Corbyn the same as weatheraction.com? They also stumbled badly. Sun wasn't bright enough to see their feet, or something.

Deano said...

Corbynwatch and Weatheraction are one and the same.

Piers doesn't like critcism by the way:

http://scienceblogs.com/stoat/2008/03/corbynwatch_shhh.php

P. Lewis said...

According to Answers.com, this seems to be the gist of the Wheeler study of PC’s method published in J Atmos Sol-Terr Phys:

Weather Action itself claims that their record demonstrates "proven skill verified by independent academic statisticians and published in scientific literature," and has been verified in Journal of Atmospheric and Solar-Terrestrial Physics.[4] The Journal study, the single academic work conducted involving Corbyn's work, was done by Dr. Dennis Wheeler of the University of Sunderland and took Weather Action predictions for the island of Britain for October 1995 to September 1997. Only predictions of gale force winds were analyzed. It uses a variety of statistical techniques to come up with success rates varying from 37% to 73% depending on the technique. 23 gales were successfully predicted, 21 gales were falsely predicted, and 18 gales occurred that were not predicted. Dr. Wheeler concluded, "that forecasts prepared by Weather Action would repay further attention. The results provide little evidence to dismiss the observed success rates as being attributable to mere chance or good fortune. Indeed the balance of evidence indicates that the system performs better than chance although it is recognized that the margin of success differs greatly between the seasons and is lowest in winter when gales are most frequent.”[5]

This is the first time I've seen anything quoted other than the Wheeler abstract. It's easy to see where the 37% comes in. I'm unclear on how the 73% might be obtained though. I could hazard a guess that in one or more months (what odds on it being an Oct, Jan or Mar?) or a season the figure reached the heady heights of 73%. Without reading the actual paper to check a few things, though, it’s difficult to know.

However, to me, “prediction” means noting that a specific event will take place at some specific future time/location. On that, I’d say 37% accurate prediction or a failure rate of ~63% might also be a conclusion. Perhaps a reading of the paper is warranted.

PS. Do you think he wants any of this written about in blogs? Because …

Corbyn now says on his website that Media or web users who wish to quote from the forecast must check what they have in mind with Piers Corbyn … and preferably use an issued summary form (below for this month) which is written to give an accurate rendering of essential points without giving away the detail which subscribers have paid for.

Gosh! We have been warned I suppose.

P. Lewis said...

Ah! About the "PS". I should have read the link to Stoat first, it seems.

James Annan said...

I've got a copy of the Wheeler paper somewhere. If it turns up I may comment in more detail.

EliRabett said...

Seems to me that an analysis of a forecast after the forecast period is no skin off any subscriber's back not that I assume anyone takes Corbyn seriously. Well maybe those subscribers.