Closed kmdesorcy closed 6 months ago
My apologies for the delay in responding, @kmdesorcy . We picked up the maintenance of corncob and are still clearing our backlog.
In any hypothesis test, you need to make sure that your full/alternative models (for both means and dispersions) are subsets of your restricted/null models.
~1
is indeed a restriction of ~A + B + C
(so your mean model is fine), but ~ D + E
is not a restriction of ~ A + B + C
. Alternatives include:
~ A + B + C
is a restriction of ~ A + B + C + D + E
.~ D + E
is a restriction of ~ A + B + C + D + E
.~ 1
is a restriction of ~ A + B + C + D + E
and so on.
In general most people are interested in differential abundance, so my general advice is that if you care about D and E, you should include them in your formula
. Including them in phi.formula
reflects a less common scientific question.
I hope this helps -- feel free to reopen if not.
Hello,
I'm new to corncob. I'm trying to create a model that tests differential abundance of a gut microbiome community based on variables A, B, and C with covariates D and E. I've come up with the following model by trying to adapt what I've read in vignettes:
example <- differentialTest(formula = ~ A + B + C, phi.formula = ~ A + B + C, formula_null = ~ 1, phi.formula_null = ~ D+ E, data = data1, test = "Wald", boot = FALSE, fdr_cutoff = 0.05)
I'm wondering if anyone with expertise using corncob can confirm that I've set my model up to test as intended? I'm a little confused about testing while controlling for effects on dispersion? If anyone could elaborate or post a resource, that would be great. I've read the cited publication.
Thank you!