Question regarding family specification and graphical model structure

[Ryan Runquist]

Hello,

I have a question regarding the correct specification of my data as I am starting to implement aster. In this experiment we started with seeds in a large field experiment. We planted seeds of an annual into the field into grid cells and followed the plants to seed-set as a measure of lifetime fitness. For the most part, a first pass, the graphical model seems relatively straightforward. However, I am struggling somewhat with the proper exponential family for germination and wanted your advice.

In the experiment, to increase the probability for success, we planted 4 seeds in each grid cell. We then surveyed for early germination, counted the number of germinants, and if there was more than one seedling we randomly removed individuals leaving only one seedling. This seedling was then followed till seed-set. Germination would then be binomial (with 0-4 possible successes), but that of course is not part of aster. I have read from the slides and some of the threads here that binomial is just the sum of n IID Bernoulli random variables. In my case would the root of my graphical model be 4 (number of seeds in each cell) with the arrow indicating a Bernoulli distribution linking to the number of germinants? Or is this incorrect because we then removed seedlings and only followed one individual so the root has to be 1? and if that is the case, what is the proper way to incorporate the germination data? 

Thank you in advance,

Ryan 

[John Stanton-Geddes]

Hi Ryan,

It is possible to start with a 'root' to your graphical model that is not 1. This is the approach I used in the analysis of my field experiment (R code is available here). 

It's also possible to include a "sampling" stage where not all the units of observation at the previous stage make it to the next. For example, I used this when counting all the seeds in only a subsample of pods (my code here. Charlie has a better example that I modified mine from in one of the tech reports):

# Set up data for aster modeling through seeds
# Subsampling seed pods to include seeds/pod in life history graph
# graph:
#        1 -> e.surv -> fecund -> seed.pods -> pods.counted(sample) -> any seeds -> seeds.counted
#         samp is binomial with sample size pods and known success probability p

That said, since your germination treatment wasn't explicitly sampled with this analysis in mind, it might not work well. 

In your case, you appear to have (at least?) two experiments:

A) Germination

B) Life history

so it seems you should just analyze them separately, using aster from the start point of an individual seedling that germinated, and a GLM for germination. Maybe not the answer you were hoping for...but probably easier!

--
You received this message because you are subscribed to the Google Groups "Aster Analysis User Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to aster-analysis-use...@googlegroups.com.
To post to this group, send email to aster-analys...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/aster-analysis-user-group/6f782dc1-da73-48e0-a359-a2b8a2104def%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.