[R-sig-ME] Grouping variables technically suitable for modeling
th|erry@onke||nx @end|ng |rom |nbo@be
Tue Nov 9 16:02:59 CET 2021
I would expect in your example that the combined effect of ID1 and ID2 will
be more or less equally split over ID1 and ID2. As this would yield a lower
penalty then attributing the effect fully to either ID1 or ID2. Hence the
random effect variances of 1|ID1/ID2 will be a lot smaller than 1|ID1 or
As ID2 defines almost the same grouping as ID1, it doesn't make sense to
include both of them in the model.
I have no reference at hand for this. Just common sense.
ir. Thierry Onkelinx
Statisticus / Statistician
Vlaamse Overheid / Government of Flanders
INSTITUUT VOOR NATUUR- EN BOSONDERZOEK / RESEARCH INSTITUTE FOR NATURE AND
Team Biometrie & Kwaliteitszorg / Team Biometrics & Quality Assurance
thierry.onkelinx using inbo.be
Havenlaan 88 bus 73, 1000 Brussel
To call in the statistician after the experiment is done may be no more
than asking him to perform a post-mortem examination: he may be able to say
what the experiment died of. ~ Sir Ronald Aylmer Fisher
The plural of anecdote is not data. ~ Roger Brinner
The combination of some data and an aching desire for an answer does not
ensure that a reasonable answer can be extracted from a given body of data.
~ John Tukey
Op di 9 nov. 2021 om 15:19 schreef Timothy MacKenzie <fswfswt using gmail.com>:
> Dear Ben,
> Thank you for sharing the references regarding my first question.
> Regarding my second question, I simply mean if we have say ID1 and ID2,
> then for ID2 to be distinguishably nested in ID1, it needs to have a
> different unique categories relative to those of ID1.
> For example, if ID1 has 120 unique categories and ID2 has 130
> unique categories nested in ID1, then the variance components for ID1 and
> ID2 are not distinguishable from each other. As a result, only one of them
> can be added as a random effect; either (1 | ID1) or (1 | |ID2), but not (1
> | ID1/ID2).
> Is this correct and is there a published reference confirming or
> disconfirming this?
> Tim M
> On Mon, Nov 8, 2021 at 7:35 PM Ben Bolker <bbolker using gmail.com> wrote:
> > This is a bit of a "how long is a piece of string" question ...
> > The "5-6 levels of a grouping variable" rule of thumb is quoted in
> > various places: a variety of those references (Gelman and Hill 2006,
> > Kéry and Royle 2015, Harrison et al 2018, Arnqvist 2020) are collected
> > by Gomes
> > (https://www.biorxiv.org/content/10.1101/2021.04.11.439357v2.full).
> > I sort of see what you mean by your second paragraph, but can you
> > give an example?
> > On 11/7/21 5:20 PM, Timothy MacKenzie wrote:
> > > Dear Experts,
> > >
> > > Apologies if this question has come up before. But I'm looking for
> > > published references that provide guidance on when one or more grouping
> > > variables that theoretically need to be random factors can also
> > > "technically" be used as random factors?
> > >
> > > For example, I have heard for a grouping variable to be technically
> > > as a random factor, it needs to have at least 10 or so unique
> > > (Any reference to confirm or disconfirm this?)
> > >
> > > For example, I have heard for two grouping variables to be technically
> > > taken as random factors, they each need to have a sufficiently
> > > number of unique categories relative to the other one. Otherwise, their
> > > variance components can't be distinguished from one another and thus
> > > one of them can be taken as random, not both (Any reference to confirm
> > > disconfirm this?)
> > >
> > > Thanks,
> > > Tim M
> > >
> > > [[alternative HTML version deleted]]
> > >
> > > _______________________________________________
> > > R-sig-mixed-models using r-project.org mailing list
> > > https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
> > >
> > _______________________________________________
> > R-sig-mixed-models using r-project.org mailing list
> > https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
> [[alternative HTML version deleted]]
> R-sig-mixed-models using r-project.org mailing list
[[alternative HTML version deleted]]
More information about the R-sig-mixed-models