[R-sig-ME] lmer under "single" nests

[Ben Bolker]

  The 2008 reference looks a little bit naive/old-fashioned to me; the
2014 paper looks more useful.

  The answers to your questions (what methods should you use etc.)
will depend on the answers to some of these questions:

* are you more interested in fixed/population-level effects or in the
variance components? (The former are easier.)
* how many groups do you have at both levels? (I think but am not not
quite sure that 'B' represents donors and 'A' represents some
higher-level grouping variable [hospital etc.]?)
* how many observations *per group*? (i.e. having 1-2 kidneys per
donor, but 4-5 measurements per kidney, is much better than having a
single kidney per donor)
* the responses are continuous/will be treated as Gaussian? That makes
things *much* easier/better than if they were binary outcomes (which
is sort of a worst-case scenario)


On Thu, Aug 11, 2022 at 8:58 AM Pierce, Steven <pierces1 using msu.edu> wrote:
>
> Elena,
>
> You have what is sometimes called "sparsely clustered" data. Below are a couple methodology papers relevant to this situation.
>
> Clarke, P. (2008). When can group level clustering be ignored? Multilevel models versus single-level models with sparse data. Journal of Epidemiology and Community Health, 62, 752-758. https://doi.org/10.1136/jech.2007.060798
>
> McNeish, D. M. (2014). Modeling sparsely clustered data: Design-based, model based, and single-level methods. Psychological Methods, 19(4), 552-563. https://doi.org/10.1037/met0000024
>
>
> Steven J. Pierce, Ph.D.
> Associate Director
> Center for Statistical Training & Consulting (CSTAT)
> Michigan State University
>
> -----Original Message-----
> From: Elena Moreno <momae1112 using gmail.com>
> Sent: Wednesday, August 10, 2022 7:44 AM
> To: r-sig-mixed-models using r-project.org
> Subject: [R-sig-ME] lmer under "single" nests
>
> Dear R-sig-mixed-models list:
>
> I first want to thank you for your attention and willingness to help people
> like me. I hope to get some light with my question:
>
> I am applying a nested model as "(1|A/B)", having two individuals per nest
> in most of the cases. However, some nests only have one individual. This is
> because I'm working with kidney transplant data: there are cases when two
> patients receive a kidney from the same donor (the donor gives both
> kidneys), but there are cases where the donor gives just one kidney (so the
> recipient doesn't share donor with anyone else).
>
> Patients have several measures of renal function (creatinine) over time.
>
> How does "lmer" handle this kind of situation when having some "single"
> (with just one individual) nests in combination with non-single nests? It
> is worth it to nest when, at most, there are only two patients per nest
> (donor)?
>
>
> If you need more details regarding the study design or even a sample of the
> data, please tell me. By the way, I am not mathematician so I find
> demonstrations difficult to understand but I am always eager and open to
> learn.
>
> Thank you very much and sorry for this naive question,
>
>
> Elena
>
>         [[alternative HTML version deleted]]
>
>
> _______________________________________________
> R-sig-mixed-models using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models