-Mensaje original-
De: Bert Gunter [mailto:gunter.ber...@gene.com]
Enviado el: jueves, 26 de enero de 2012 21:20
Para: Rubén Roa
CC: Ben Bolker; Frank Harrell
Asunto: Re: [R] How do I compare 47 GLM models with 1 to 5 interactions and
unique combinations?
On Wed, Jan 25, 2012 at 11:39
Ruben, I'm not sure you are understanding the ramifications of what Bert
said. In addition you are making several assumptions implicitly:
1. model selection is needed (vs. fitting the full model and using
shrinkage)
2. model selection works in the absence of shrinkage
3. model selection can find
-Mensaje original-
De: r-help-boun...@r-project.org [mailto:r-help-boun...@r-project.org] En
nombre de Frank Harrell
Enviado el: viernes, 27 de enero de 2012 14:28
Para: r-help@r-project.org
Asunto: Re: [R] How do I compare 47 GLM models with 1 to 5 interactions and
unique combinations?
Ruben you are mistaken on every single point. But I see it's not worth
continuing this discussion.
Frank
Rubén Roa wrote
-Mensaje original-
De: r-help-bounces@ [mailto:r-help-bounces@] En nombre de Frank Harrell
Enviado el: viernes, 27 de enero de 2012 14:28
Para: r-help@
Asunto:
What variables to consider adding and when to stop adding them depends greatly
upon what question(s) you are trying to answer and the science behind your data.
Are you trying to create a model to predict your outcome for future predictors?
How precise of predictions are needed?
Are you trying
Thank you everyone for your dedication to improving 'R' - its function to
statistical analysis and comments.
I have now 48 models (unique combinations of 1 to 6 variables) and have put
them into a list and gained the results for all models. Below is a sample of
my script results:
m$model48 -
To pretend that AIC solves this problem is to ignore that AIC is just a
restatement of P-values.
Frank
Rubén Roa wrote
I think we have gone through this before.
First, the destruction of all aspects of statistical inference is not at
stake, Frank Harrell's post notwithstanding.
Second,
I ask the question about when to stop adding another variable even though it
lowers the AIC because each time I add a variable the AIC is lower. How do I
know when the model is a good fit? When to stop adding variables, keeping
the model simple?
Thanks, J
--
View this message in context:
Simple question. 8 million pages in the statistical literature of
answers. What, indeed, is the secret to life?
Post on a statistical help list (e.g. stats.stackexchange.com). This
has almost nothing to do with R. Be prepared for an onslaught of often
conflicting wisdom.
-- Bert
On Thu, Jan 26,
Le mardi 24 janvier 2012 à 20:41 -0800, Jhope a écrit :
Hi R-listers,
I have developed 47 GLM models with different combinations of interactions
from 1 variable to 5 variables. I have manually made each model separately
and put them into individual tables (organized by the number of
A more 'manual' way to do it is this.
If you have all your models named properly and well organized, say Model1,
Model2, ..., Model 47, with a slot with the AIC (glm produces a list with one
component named 'aic') then something like this:
x - matrix(0,1081,3)
x[,1:2] - t(combn(47,2))
for(i in
If you are trying to destroy all aspects of statistical inference this is a
good way to go. This is also a good way to ignore the subject matter in
driving model selection.
Frank
Jhope wrote
Hi R-listers,
I have developed 47 GLM models with different combinations of interactions
from 1
Rubén Roa rroa at azti.es writes:
A more 'manual' way to do it is this.
If you have all your models named properly and well organized, say
Model1, Model2, ..., Model 47, with a slot with the AIC (glm
produces a list with one component named 'aic') then something like
this:
x -
I think we have gone through this before.
First, the destruction of all aspects of statistical inference is not at stake,
Frank Harrell's post notwithstanding.
Second, checking all pairs is a way to see for _all pairs_ which model
outcompetes which in terms of predictive ability by -2AIC or
14 matches
Mail list logo
