Dear Martijn,
You are right that the change in the summary defaults was not a sensible
thing to do, and I'll change it back. However in the mean time you can
use 1.7-17 with summary(...,freq=FALSE), when you have random effects in
the model (it really shouldn't make a statistically meaningful
Hi Simon,
Thanks for your reply. That is very helpful.
However, in the logistic regression example, there also appears to be
a large difference in the p-values and estimates when comparing
summary(model) # mgcv 1.7-11
summary(model,freq=F) # mgcv 1.7-17
These should be the same, right? But why
Hi Martijn,
The issue in this case is that the estimates for the terms are rather
different (not just the p-values). I think that the difference is down
to this change made in version 1.7-13 (from changeLog)...
* re smooths are no longer subject to side constraint under nesting
(since this
Hi Martijn,
The negative edf from fREML was a matrix indexing bug triggered by the
re term (other fREML discrepancies followed from this). Fixed for
1.7-18. Thanks for reporting this, and the supporting offline info!
best,
Simon
On 02/06/12 23:17, Simon Wood wrote:
the fREML results must
Dear useRs,
I ran an additional analysis with another dataset and (logistic
regression), but also in this case I obtain very different p-values
comparing version 1.7-11 and 1.7-17. Any ideas about this? Which is
the good one?
Below, I'll first show the model based on 1.7-11, then of version
Dear useRs, Simon,
@Simon: I'll send the data offline.
The command summary(...,freq=F) yields the same result as before. Why
did the default change from freq=F in version 1.7-13 to freq=T in
version 1.7-17? Especially since the original default appeared to be
better.
I noticed that some of my
Dear useRs,
I've ran some additional analyses (see below), which strongly suggest
the standard errors of the bam (and gam) function are much too low in
mgcv version 1.7-17, at least when including an s(X,bs=re) term.
Until this issue has been clarified, it's perhaps best to use an older
version
Dear useRs,
I reran an analysis with bam (mgcv, version 1.7-17) originally
conducted using an older version of bam (mgcv, version 1.7-11) and
this resulted in the same estimates, but much lower standard errors
(in some cases 20 times as low) and lower p-values. This obviously
results in a larger
the fREML results must be a bug. I'd modified the bam covariance and edf
code in 1.7-17 and it looks like I must have messed something up. Any
chance you could send me the data off line?
Also can you try summary(...,freq=FALSE) under 1.7-17 and see if you get
the same results as before?
On
9 matches
Mail list logo