[CF-metadata] how to represent a non-standard error

From: Jonathan Gregory <j.m.gregory>
Date: Wed, 10 Jul 2013 10:41:48 +0100

Dear Nan and Randy

Thanks for the descriptions you forwarded. Neither of these really seems to
answer the question quantitatively, however. They don't say how to interpret
the error. If I have a value X with an error Y, what can I do quantitatively
with Y? If it's a standard error, and I assume that errors are normally
distributed, I could infer there is a 95% chance that the true value is between
X-2Y and X+2Y, for instance. That's because a standard error is an estimate of
the standard deviation of values. Would you use the errors that you deal with
like this? If so, we can call them standard errors. If not, how would you (or
the users of your data) use them? If we knew that, we would be better able to
decide how to describe them.

Best wishes

Jonathan
Received on Wed Jul 10 2013 - 03:41:48 BST