"Konrad Den Ende" <[EMAIL PROTECTED]> wrote in message news:<[EMAIL PROTECTED]>...
> >> L = \sum_{i=1}^b \sum_{j=1}^k (y_ij - \beta_i - \mu_j)^2
> >
> > To narrow it down, could you say what you think the derivative of the
> > above loss function is with respect to beta_i?
>
> The partial derivatives i get are:
> dL/d(beta_i) = -2 \sum_{i=1}^b \sum_{j=1}^k (y_ij - \beta_i - \mu_j)
Here's your problem right here. This is not correct.
> dL/d(mu_j) = -2 \sum_{i=1}^b \sum_{j=1}^k (y_ij - \beta_i - \mu_j)
As I pointed out, this is superfluous, because of the symmetry of
beta and mu in the model.
> > (This is homework, right?)
>
> No
Okay, thanks. It helps to know that, because if it was homework
I would tend to give a slightly different form of help (though
with the problem you have above, I'd still have pointed it out
the same way).
> it's actually a question in a book, "Mathematical Statistics And It's
> Applications" by Larsen and Marx. It's question 13.2.12 on page 682 if you
> happen to have it in your bookshelf.
Since many homework questions come from textbooks, me finding it in a
text doesn't demonstrate it either way. But I am happy to take your
word for it.
Glen
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
. http://jse.stat.ncsu.edu/ .
=================================================================
