[R-sig-eco] Appropriate adonis call for time-series data?

[Lawrence David]

Hi everyone,

I’m hoping some of you may be able to clarify how to use adonis to analyze time-series data of microbial communities.

I’ve collected community data from different individuals after the same perturbation.

I’d like to determine whether individual identity and/or time have effects on the sampled communities.

My plan for analyzing these data is to consider ‘time’ a fixed effect and ‘individual’ a random effect.

If this sounds reasonable, what might be the preferred approach for implementing this model in adonis? I am considering the following model, based on reading the vegan FAQ and these forums: adonis(dist_matrix ~ Individual + Time, strata=Individual)

I should mention that I’ve already tried this and found some curious things. First, I’ve run:

————————
adonis(formula = vegdist(data,method="bray") ~ Individual+Time, permutations=10000, strata=Individual)
Terms added sequentially (first to last)
                        Df SumsOfSqs MeanSqs F.Model      R2 Pr(>F)
Individual            12    4.5076 0.37563  1.2296 0.36641  2e-04 ***
Time  1    1.6845 1.68446  5.5139 0.13693  2e-04 ***
Residuals               20    6.1099 0.30549         0.49666
Total                   33   12.3019                 1.00000
————————

This seems surprising, as should I be able to detect a significant association with Individual, given that I’ve stratified on that variable?

I’ve also tried running adonis on just Individual alone: adonis(formula = vegdist(data, method = "bray") ~ Individual, permutations=10000, strata=Individual)

This is also a bit puzzling, as I get different results on my two workstations. According to my workstation install (R version 3.1.2 (2014-10-31); vegan 2.2-1), I get:

————————
Call:
adonis(formula = vegdist(data, method="bray") ~ Individual, permutations=10000, strata=Individual)

Blocks:  strata
Permutation: free
Number of permutations: 10000

Terms added sequentially (first to last)

            Df SumsOfSqs MeanSqs F.Model      R2    Pr(>F)
Individual   12    4.5076 0.37563  1.0121 0.36641 9.999e-05 ***
Residuals    21    7.7943 0.37116         0.63359
Total        33   12.3019                 1.00000
————————

But, on my laptop (R version 3.0.3 (2014-03-06); vegan 2.0-9), the same call gives nearly the same results, but with different significance: Pr(>F)=1. Is strata not restricting permutations the way I want it to in the newer version of vegan?

So, to summarize, my questions are:

1) Am I using the right model design to test for an impact of time on a community sampled over time in subjects (i.e. adonis(dist_matrix ~Individual+Time,strata=Individual)
2) Is this a valid test for whether ‘Individual’ contributes to community structure?
3) Why might I be getting different results for running ‘Individual' as a single factor and a stratifying variable at the same time?

Thank you in advance for any help you can lend!

- Lawrence David