| |
 |
|
|
Science Forum Index » Space - Consult Forum » bootstrap over uncertain measurements...
Page 1 of 1
|
| Author |
Message |
| Mark Wexler... |
 Posted: Thu May 15, 2008 1:52 am |
|
|
|
Guest
|
Let's say I do an experiment on N subjects, and for each subject I
have a measurement m_1, ..., m_N. To do a bootstrap resample, I choose
N measurements, with replacement.
OK, now let's say that for each subject, I actually have a bunch of
complicated measurements, from which I can calculate the most probable
m_i. But in reality, for each subject I have a distribution for his or
her m_i, let's call it D_i. These distributions are not the same: for
some subjects D_i are sharply peaked, so that I know m_i with high
precision, while for others the D_i are highly spread out, so that the
corresponding m_i are known with low precision.
So how do I do a bootstrap resample on this kind of data? I can think
of three possibilities:
1. As before, I choose N subjects with replacement, and for each
subject i, I choose his or her most likely m_i.
2. I keep my original N subjects, but for each subject I choose an
m_i* from his or her individual distribution D_i.
3. A combination: I choose N random subjects (with replacement), and
for each subject i I choose a random m_i from his or her distribution
D_i.
I anyone has any insight on this question, I'd really appreciate it.
Mark Wexler |
|
|
| Back to top |
|
| |
|
Page 1 of 1
All times are GMT - 5 Hours
The time now is Fri Dec 05, 2008 7:48 am
|
|