Prev: Group theory, semidirect products
Next: Derivative of a Fourier transform
From: lorenzo123 on 13 Jun 2010 15:02
I have a two sets of N paired real numbers,
P=predictions
O=observed values

I am computing the "average prediction error" as

ape=sum(abs(P_i - O_i)) / sum(abs(O_i)), 1<=i<=N

How can I calculate a confidence interval on ape?

Thanks a lot!
Lorenzo
From: Chip Eastham on 13 Jun 2010 21:56
On Jun 13, 7:02 pm, lorenzo123 <suruenf...(a)gmail.com> wrote:
> I have a two sets of N paired real numbers,
> P=predictions
> O=observed values
>
> I am computing the "average prediction error" as
>
> ape=sum(abs(P_i - O_i)) / sum(abs(O_i)), 1<=i<=N
>
> How can I calculate a confidence interval on ape?
>
> Thanks a lot!
> Lorenzo

Not without some assumption about the distribution
of errors, possibly obtained from information about
the prediction model and the sample and/or population
distributions.

regards, chip
From: lorenzo123 on 14 Jun 2010 02:39
Assume normal distribution.
 | 
Pages: 1
Prev: Group theory, semidirect products
Next: Derivative of a Fourier transform