Gang Wrote:
-------------------------------------------------------
> > lme(values ~ time*treatment +
> ratings,data=lme_data,random = ~1 + ratings |
> subj)
>
> > I get the following error message:
> > nlminb problem, convergence error code = 1
> > message = iteration limit reached without
> convergence (10)
>
> This is basically what 3dLME does except for a
> couple of differences (see below). Basically it
> failed to converge at that voxel for some
> numerical reason. Try a different voxel, and see
> if it converges.
Great. I tried a different voxel and it worked. I tried it again with the same voxel giving the error message and it worked. It might have been a formatting issue.
>
> > lme(values ~ time*treatment +
> ratings,data=lme_data,random = ~1 + ratings |
> subj)
>
> The reasons you had some different results from
> 3dLME are:
>
> 1) A different dummy coding scheme is adopted in
> 3dLME than the default in the nlme package. I use
> "deviation coding" (contr.sum in R) in 3dLME while
> the simple "dummy coding" (contr.treatment in R)
> is used in the nlme package. The consequence of
> the different coding strategies is that the
> intercept is interpreted totally different between
> the two.
>
> 2) The default for the type of sums of squares in
> 3dLME is "marginal" (or type 3) while the default
> in the nlme package is "sequential" (or type 1).
> And this is why you see the different F-values
> (and p-values as well) between your analyses
> except for the last term (interaction). See more
> discussion about the issue here:
>
> [
afni.nimh.nih.gov]
This is very helpful. Selecting the two options of marginal in anova.lme and contr.treatment for the options for contrasts, I get the same results using lme as I do 3dLME.
Thanks again.
Michael