[R-sig-ME] nesting hierarchy in glmer

[Thierry Onkelinx]

Dear Hariet Rose,

It looks like you scenario's (temperature:dose) were too extreme since all
seedlings died in some scenario's. Keep that in mind when designing your
next experiment.

First of all adding room to both the fixed effect as the random effect is
not OK in case of categorical variables:
https://rpubs.com/INBOstats/both_fixed_random

To salvage as much as possible I'd remove the scenario's in which all
seedlings died. Then fit one model:

glmer(survivorship ~ pops * scenario + (1|room/id), family = binomial, data
= all.relevant.data)

Best regards,


ir. Thierry Onkelinx
Instituut voor natuur- en bosonderzoek / Research Institute for Nature and
Forest
team Biometrie & Kwaliteitszorg / team Biometrics & Quality Assurance
Kliniekstraat 25
1070 Anderlecht
Belgium

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

2016-05-29 18:24 GMT+02:00 Hariet Rose <harietrose123 op gmail.com>:

> Hi, I am a newbie at posting on this forum and relatively new to R.  My
> question is about nesting in glmer and examples on the forums do not
> correspond to my experiment, especially as I think I could run my design as
> two separate analyses.
>
>
> I have a glasshouse complex study where I have six independent rooms. Three
> rooms were randomly chosen to be set at the hot temperature for the
> duration of the experiment, and another three rooms were randomly chosen
> and set at a cold temperature for the duration of the experiment (13
> weeks).  This is why I think I could run two separate models
>
> .
>
> [There were other rooms that I didn’t use although they were ‘available’ to
> be chosen randomly].
>
>
>  For my question, I want to know if two populations (pops = pop1, pop2) of
> seedlings respond in different ways to 3 water dosages under the
> temperature regimes. The rooms were stocked evenly with seedlings from two
> different populations of the same species; each seedling in each room
> received only 1 of 3 water doses; local, wet or dry.  I measured
> survivorship of seedlings at week 13 and the response was 1 = alive or 0 =
> dead.  In the models below “id” is the code I have given to all plants that
> received the same treatment within a room (e.g.id 1=  replicate 1, all
> plants from pop1 in room1 (hot)  that received the local water treatment)..
>
>
> As all seedlings died when the following treatments were applied;  hot +
> dry, hot +local and cool + dry, I have a very skewed response.  This
> suggests to me that some model simplification may be possible...
>
>
> Because the two temperatures could be seen as independent experiments and
> because of the skewed response I am thinking it would be simpler to run two
> separate models, one for the cool rooms and another for the hot rooms;
>
>
> R version 3.1.3 (2015-03-09), library(lme4)
>
> coolmodel <- glmer(survivorship ~ room*pops*dose * (1|room/id), family
> = binomial, data = mydata1)
>
>  mydata1 is a matrix of cool room data only and the samples from the
> dry dose treatment have been removed (i.e., both pops of dry dosed
> plants died).
>
> hotmodel <- glmer(survivorship ~ room*pops * (1|room/id), family =
> binomial, data = mydata2)
>
> mydata2 is a matrix of hot room data only and the samples from the dry
> and local have been removed (i.e., both pops of dry and local dosed
> plants died)
>
> in the above model, because there were only survivors for one dose
> (wet) I have not specified the fixed effect of dose.
>
> Question 1. Is the above approach correct? I am concerned that the
> nesting of id may be incorrect.
>
> For the combined cool and hot room data set, I have used this model
>
> bothtempsmodel <- glmer(survivorship ~ temp*pops*dose *
> (1|temp/room/id), family = binomial, data = mydata)
>
> where mydata has all samples from all temps, rooms, and treatments –
> i.e., no data have been removed.
>
>
> Question 2. In the bothtempsmodel, if I have explained the nesting and
> everything else correctly, I end up with many significant two and the
> three way interaction/s whereas coolmodel and hotmodel have outputs
> that are simplified.
> Would a strategy be to specify the full model (bothtempsmodel) and
> then go on to explain and implement the simplified coolmodel and
> hotmodel, with the removal of data as explained above.
>
>  I can post the outputs if anyone is interested..
>
>  Thank you very much for your time.. :) hariet-rose
>
>         [[alternative HTML version deleted]]
>
> _______________________________________________
> R-sig-mixed-models op r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

	[[alternative HTML version deleted]]