Daniel Ezra Johnson johnson4 at babel.ling.upenn.edu writes:
1) Yes, I have tweaked the data to show as clearly as I can that this is a
bug, that a tiny change in initial conditions causes the collapse of a
reasonable 'parameter' estimate.
I would not call this a bug, since this is related
I am fitting models to the responses to a questionnaire that has
seven yes/no questions (Item). For each combination of Subject and
Item, the variable Response is coded as 0 or 1.
I want to include random effects for both Subject and Item. While I
understand that the datasets are fairly small,
Daniel Ezra Johnson johnson4 at babel.ling.upenn.edu writes:
I am fitting models to the responses to a questionnaire that has
seven yes/no questions (Item). For each combination of Subject and
Item, the variable Response is coded as 0 or 1.
I want to include random effects for both
According to pp 197-198 of MASS 4, the Hauck-Donner phenomenon refers
to cases when the Wald approximations and the likelihood ratio tests
have different p values because of the former underestimating the the
change in log-likelihood on setting \beta_i = 0.
This seems quite different to me than
From: Andrew Robinson A.Robinson_at_ms.unimelb.edu.au
Date: Mon 01 Jan 2007 - 19:19:29 GMT
I tried an earlier version of R, on a different platform, and got quite
different results. Sadly, the *earlier* results are the ones that make
sense.
Andrew, I tried installing R 2.3.1, it seemed to
Good observation, Daniel. I had not picked up the different default
fitting method.
Cheers
Andrew
On Mon, Jan 01, 2007 at 08:59:55PM -0500, Daniel Ezra Johnson wrote:
From: Andrew Robinson A.Robinson_at_ms.unimelb.edu.au
Date: Mon 01 Jan 2007 - 19:19:29 GMT
I tried an earlier version
On Sun, 31 Dec 2006, at 05:50 PM, Daniel Ezra Johnson
[EMAIL PROTECTED] wrote:
Gregor,
Thanks for your replies.
1) Yes, I have tweaked the data to show as clearly as I can that
this is a
bug, that a tiny change in initial conditions causes the collapse of a
reasonable 'parameter'
I am fitting models to the responses to a questionnaire that has
seven yes/no questions (Item). For each combination of Subject and
Item, the variable Response is coded as 0 or 1.
I want to include random effects for both Subject and Item. While I
understand that the datasets are fairly
Daniel Ezra Johnson johnson4 at babel.ling.upenn.edu writes:
...
If one compares the random effect estimates, in fact, one sees that
they are in the correct proportion, with the expected signs. They are
just approximately eight orders of magnitude too small. Is this a bug?
...
BLUPs are
If one compares the random effect estimates, in fact, one sees that
they are in the correct proportion, with the expected signs. They are
just approximately eight orders of magnitude too small. Is this a bug?
BLUPs are essentially shrinkage estimates, where shrinkage is
determined with
I've found a way to make this problem, if it's not a bug, more clear.
I've taken my original data set A and simply doubled it with
AA-rbind(A,A).
Doing so, instead of this:
Random effects: # A
Groups NameVariance Std.Dev.
Subject (Intercept) 1.63e+00 1.28e+00
Item(Intercept)
I'm not sure that shrinkage is the answer, in this case. I observed a
similar problem with the gamma distribution, which I mentioned here:
http://tolstoy.newcastle.edu.au/R/e2/help/06/12/6903.html
Since there hasn't been any discussion, I'm starting to think that it
is a bug.
Andrew
On Sun,
Daniel Ezra Johnson johnson4 at babel.ling.upenn.edu writes:
...
More broadly, is it hopeless to analyze this data in this manner, or
else, what should I try doing differently? It would be very useful to
be able to have reliable estimates of random effect sizes, even when
they are rather
Daniel Ezra Johnson johnson4 at babel.ling.upenn.edu writes:
If one compares the random effect estimates, in fact, one sees that
they are in the correct proportion, with the expected signs. They are
just approximately eight orders of magnitude too small. Is this a bug?
BLUPs are
Gregor,
Thanks for your replies.
1) Yes, I have tweaked the data to show as clearly as I can that this is a
bug, that a tiny change in initial conditions causes the collapse of a
reasonable 'parameter' estimate.
2) mcmcsamp() does not work (currently) for binomial fitted models.
3) This is
15 matches
Mail list logo
