ANCOM-BC formula when using multiple metadata columns

[colinbrislawn (Colin J Brislawn)]

@salias

So the intercept (i.e., reference) is taking into account both conditions. So:

That's right.

I think the Qiime2 plugin selects a baseline group to test against. Maybe by what is first in the metadata?
So for var1, A becomes baseline and for var2, Y becomes baseline.

@salias

For example, I obtain three LFC columns: var1::B, var1::C and var2::N. Let's take var1::C.

var1::A is missing, because it's the baseline
var2::Y is missing, because it's the baseline

You can also set a baseline explicitly using the plugin. There may be even more options in the R package.

@salias

And the other thing I don't know I understood correctly is that there may be a sample both in reference and comparing group.

This does not sound right...

I think results for all samples are reported, including those in the baseline cohort.

---

@salias

What is the effect or variable order? In my example, wouldn't I get the same three LFC columns when using var1 + var2 or var2 + var1?

I tried to find official R docs on this and could not.

Some tests are invariant to order, so their formulas are too. Easy!
Some packages implement tests in which the order of variables matters, like vegan::adonis().

Try it out for yourself!