[R-sig-ME] lmer multivariate longitudinal model
joerg.luedicke at gmail.com
Thu Oct 4 23:13:03 CEST 2012
Before checking out other packages, I would first try to figure out
the reason(s) for the convergence issues. More often than not the
reason is just some trivial issue with the data, but could of course
also be due to analyzing some complex data, a problem in the model
set-up itself etc. Did you fit separate models for each of your
outcome variables? Did you experience any convergence problems there?
Did the results made sense, based on your knowledge of the data?
Re multilevel/mixed-effects modeling in R, if you did not have
already, maybe have a look at this page:
On Thu, Oct 4, 2012 at 3:36 PM, basile pinsard
<basile.pinsard at imed.jussieu.fr> wrote:
> Thanks for your answer.
> The outcome variables are correlations of measures from different regions of
> interest in images, which enables them to be regressed by common variables
> (that I named 'other_confounds' in the formula). The data is the original
> data reshaped (with melt from reshape package) so that 'variable' contains
> the name of the outcome and value it's value.
> About the non-convergence, being new to R, I was myself surprised that lmer
> returns result object event if this results does not makes sense, that is
> why I do asked about this warning.
> The second warning is independent of the convergence problem that have to be
> fixed and which is mainly due to the fact of having that many variables into
> the model.
> This second warning seems to say that the control parameters are not taken
> into account whatsoever.
> I have been looking at SabeR and exploring MCMCglmm today as multivariate
> support seems built-in, but there are so many packages that it is difficult
> as a newbie to make a choice.
> On Thu, Oct 4, 2012 at 9:18 PM, Joerg Luedicke <joerg.luedicke at gmail.com>
>> It is not clear to me how your data structure looks like, what your
>> outcome variables are, whether you got reasonable results from
>> separate model fits, what exactly you did by reshaping, and so on.
>> Thus I will only throw in some 2c regarding the two questions at the
>> bottom of your post (see below):
>> On Wed, Oct 3, 2012 at 3:45 AM, basile pinsard
>> <basile.pinsard at imed.jussieu.fr> wrote:
>> > -Does the first warning implies that the outputted estimate is wrong?
>> The warning message means what it says: your model did not converge.
>> The results in the mer object are (I would think) the estimates from
>> the last iteration. However, since your model did not converge, these
>> results are at least untrustworthy.
>> > -Why are the control arguments not taken into account? (sessionInfo
>> > outputs
>> > lme4_0.99999911-0 as version) Does it makes any sense trying to change
>> > these to remove the warning?
>> The fact that your model did not converge can have many reasons,
>> including data management issues. So your question should rather be
>> "Why is my model not converging?" instead of trying to "remove the
>> warning"! If you are unsure about what convergence means in this
>> context you should probably do some basic read-up.
>> R-sig-mixed-models at r-project.org mailing list
More information about the R-sig-mixed-models