I Have Been Wrong 3,388 Times. Here Is The Breakdown.

ITHACA, N.Y. — Since 4 September 2006 I have issued 5,214 forecasts for this newspaper. By my own scoring rule I have been wrong on 3,388 of them.

That is a 65 per cent error rate, and I publish it every Friday, and every few months somebody writes in to ask why I still have a job.

What "wrong" means here

It means the observed conditions fell outside the bounds I stated. Not outside a generous interpretation of them. Outside the numbers I printed.

Most published forecast accuracy figures are not calculated this way, and the ones that report 90 per cent are usually scoring a much easier question — did it rain at all, within a wide window, at any station in a region.

My rule is harsh on purpose. A forecast that cannot be scored strictly is not a forecast, it is a mood, and this desk has enough trouble with moods already.

March and April are my worst months and always have been. The spring transition over the Finger Lakes is genuinely hard and I am genuinely bad at it.

The 2019 spike

In March and April of 2019 I was wrong on nine out of ten forecasts.

The reason is not interesting and I am going to print it anyway, because the interesting reasons are the ones people invent when you do not give them the boring one. My anemometer had been mounted, during roof repairs the previous autumn, four metres from a new plant enclosure, and it spent two months reading a wind that only existed within four metres of itself.

I did not notice for nine weeks. A reader noticed, from her garden, eleven kilometres away, and wrote to say my wind speeds had gone strange. She was right and I have sent her a Christmas card every year since.

The change I am making

The dispersal index — this desk's own nine-point measure, which I introduced in 2011 over two editors' objections — moves out of the forecast box on Monday.

A reader in Binghamton wrote to us this month with nine years of her own data showing a correlation of 0.03 between the index and anything at all in her life, and made an argument I have not been able to answer: that a number printed in the weather box becomes a weather number in the reader's head, whatever the disclaimer under it says.

She is right. Disclaiming a thing does not undo the effect of where you put it. From Monday the index has its own panel, its own heading, and her coefficient printed underneath it with her name on it.

Dr. Ferreira-Ludlow, who helped me design the verification rule, thinks I should retire the index altogether and told me so in writing. Her objection, printed at her request: "You have built a measure with no demonstrated relationship to anything and moved it four inches. Four inches is not an answer."

She may be right too. I am going to see how Monday goes.