Closed adw96 closed 2 years ago
This would be useful! I have two specific questions related to covariates for DivNet and betta. It's probably easiest to explain it in the context of my research questions.
I'm testing two similar hypotheses:
First question: Do the covariates in DivNet and the betta model need to be the same? I'm excluding infants who received antibiotics, were birthed via C-section, or reported recent illness, so those variables aren't of concern. However, infant feeding status (breastfeeding vs formula) is a strong predictor of infant GM diversity. Since infant feeding doesn't confound (it may rather partially mediate...) the relationship between depression/stress during pregnancy and infant GM diversity, I don't think it's necessary to adjust for it in the betta hypothesis testing model. But, it's a strong predictor of GM diversity, so should I include it in the DivNet models to estimate Shannon diversity? Perhaps it would increase precision?
Second question: Should I run two separate DivNet models for each hypothesis to be tested using betta? (The first DivNet model with depression and the second with perceived stress instead, with SES (and possibly infant feeding) as a covariate in both)
Hi @ejwatson – I think it would be helpful to have a little more information here. Do you have technical replicate observations? If not, the simplest approach may be to use breakaway to estimate diversity in each sample and then use breakaway::betta
to run your regression. If you do have technical replicates, I'd recommend using DivNet with a design matrix that groups technical replicates from the same specimen together (but doesn't include any other covariates). There's an example of how to do this in the DivNet beta diversity vignette.
Feel free to reopen @ejwatson, but closing since we haven't heard from you in a while 🙂
... in
DivNet
andbetta()
would be very useful!