[R-sig-ME] lmer(): Higher weight in case of more measurements

[Bradley Carlson]

Despite searching for previous discussions of the issue (which are often
wrestling with more specific questions than the one stated here by
Susanne), I too am confused about how lmer handles it when there is
variation in size among different groups, clusters, or whatever we want to
call levels of a random effect. Maybe I'm using the wrong search terms, but
my efforts seem to reveal that many people are confused by this and - with
the explosion of mixed models in research - are in need of guidance to
prevent serious but simple errors in model specification from being
published. Is there anyone who can give some straightforward guidance on
the issues?

What is the best practice(s) for estimating the effect of an independent
variable in a basic LMM (Y~X + (1|Group)) where the number of data points
varies among the groups?

Thank you in advance to whoever can shed some light!

On Fri, May 15, 2015 at 8:28 AM, Susanne Susanne <susanne.stat at gmx.de>
wrote:

> Dear Mr Bates and Mr Bolker, Dear R-list,
>
> I have a question regarding the lme4 package and would be very thankful if
> you could help me.
> My data consists of several clusters with repeated measurements x.
> Therefore, I want to use the lmer() function to regress the data using a
> mixed model:
>
> lmer(y ~ x + (x | cluster).
>
> I want to weight clusters, which have more measurements than others,
> slightly higher, as they provide more information.
>
> Is this done automatically in the lmer() function or should I do it
> manually by myself? And how could this be realized?
>
> I tried out small examples with fictitious data (and quite a few with real
> data) but it seems impossible for me to decide whether lmer() is weighting
> these clusters higher.
> If I weight some clusters higher with the “weights=” argument, it seems to
> matter how many measurements the cluster has. If I set default weights it
> doesn't seem to matter.
>
> Many thanks in advance,
>
> Yours sincerely,
>
> Susanne
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>



-- 

Bradley Evan Carlson
Assistant Professor of Biology
Wabash College, Crawfordsville IN

Email: *carlsonb at wabash.edu* <+carlsonb at wabash.edu>
Website: https://sites.google.com/site/bradleyecarlson/home

	[[alternative HTML version deleted]]