Could you please clarify the appropriate use of beta-group-significance and q2-longitudinal for my data? My data consists of sites (a, b, and modified sites aa, bb) as well as time differences (sampled at time 0, 2, 5 and 10 weeks, but modified sites only 2, 5, and 10). I was hoping to look at the differences between the sites across the time points. I have used beta-group-significance and its pairwise comparisons, but almost all pairs were significantly different (between time points from the same site, as well as between sites during the same point time), so I am not certain whether Iām using this test correctly. For example, if I would like to look at the differences between site A at t2 vs site B at t2, am I supposed to use beta-group significance? Is it also correct to look at the pairwise comparisons between site A t2 vs site A t6 using beta-group-significance, or is q2-longitudinal more appropriate for differences across time? What I am trying to determine is whether the community between the sites become more similar as time progresses. Thank you
Hi @jng,
Theoretically PERMANOVA is not supposed to be used on longitudinal data, due to temporal autocorrelation issues (I believe this is discussed in one of Marti Anderson's first couple papers on PERMANOVA, so you should check there for more details).
Sounds like you are not necessarily even interested in the changes over time, though; it sounds like you are most interested in the between-group differences at each individual timepoint:
So you could just filter to each timepoint and compare groups at each individual timepoint with the hypothesis that effect size (R2) will reduce over time, and maybe significance would disappear...
pairwise-distances in q2-longitudinal would also be a good test to use, but it asks a different question; basically, is one site changing at a different rate than another? BTW, you can also use first-distances followed by linear-mixed-effects to ask this same question, but over all 4 timepoints rather than looking at individual pairs (e.g., if you expect a gradual change across time in each site)
Thank you for your help and your thorough description of how I should proceed with my analysis! I will filter my comparisons to each time point using beta-group-significance, followed by first-distances and linear-mixed effects to determine whether my sites are changing across time.
