On Thu, Apr 14, 2011 at 09:19:47PM -0700, Bill Unruh wrote:
I will try to write this up and send it to you (have a colloquium to give
tomorrow, so it might take a day or two).
Ok, thanks.
But the fact that the current
precedure leads to and instability and in particular can badly underestimate
the value of the variances meand that I think for now it would be a good idea
to go back to the original and to look at this problem more deeply. I do agree
that using sigma is better, but the procedure at present simply does not work,
and I cannot see a simple fix. Anything simple seems to leave the system
unstable.
Even the very latest code which uses the unweighted variance? I
haven't seen any problem with it yet.
Skew and hi/lo are from the sd of the slope, which I believe wasn't
affected by this problem, only variance and sd of the intercept.
Since the variance of the slope is directly related to the variance of the
values, it is also affected.
From the statistics log before the fail:
==========================================================================
Std dev'n Est offset Offset sd Diff freq Est skew Stress Ns Bs Nr
==========================================================================
1.031e-143 5.934e-07 5.155e-144 1.998e-08 2.941e-07 3.4e-01 6 2 3
2.644e-145 -1.665e-07 1.182e-145 5.702e-09 5.585e-08 3.8e-02 5 0 4
4.370e-147 1.660e-08 1.784e-147 9.544e-10 1.851e-08 9.5e-03 6 0 5
1.127e-148 -3.025e-07 4.259e-149 -7.828e-09 1.919e-08 4.3e-01 7 0 5
1.012e-150 -1.086e-07 5.061e-151 -8.074e-09 5.962e-08 3.2e-01 8 4 3
7.136e-153 2.590e-08 3.191e-153 -3.025e-09 1.282e-08 4.4e-02 5 0 4
7.090e-155 9.980e-08 2.894e-155 2.693e-09 9.664e-09 2.2e-01 6 0 5
4.220e-157 -2.490e-08 1.595e-157 -6.246e-10 3.986e-09 1.0e-01 7 0 5
1.725e-158 6.033e-07 8.624e-159 2.340e-08 2.093e-07 5.9e+00 8 4 3
4.370e-160 -6.380e-08 1.954e-160 1.379e-08 4.149e-08 4.6e-02 5 0 4
-nan -nan -nan -nan -nan nan 6 0 10