I have a data set of an insect community composition (17 insect species, raw abundance data, sampled 5 times over 60 days within 24 tanks( 4 replicates). There were 3 treatments( free fish, caged fish, absent fish) involved with 2 levels (open and closed tanks) for each treatment.
And my question is does community composition and diversity change over time? And if so, are these changes different under different treatments?
I am considering to use bray-curtis index(BCI) to address this question. However by looking BCI up, it seems that BCI calculates coeffiecients between different sites(spatial distances). If that's true, my dataset doesn't have sites. I was wondering if it's possible that replicates or treatments can be considered at sites (spatial distance)?
Or is PERMANOVA a better way to answer this question?
(a preview of what my dataset looks like in is in the jpeg file)
Dear Idriss,
Thanks for your response!
I dont understand what you meant by non-independent data points.
I have concluded that a PERMANOVA would be a good choice to answer my question because it doesnt take into account the distribution of my data. I used a shapiro.test to test for normality and that was ok, but my variances were not equal.
My next question is picking the best code for it. Because i have to take into account change in time under different treatments and that i possibly have a nested effect (different levels of opened and closed tanks), i'm confused on which symbols to use to address my nestedness as i've read that i can just use the term "strata= group" or just put "/" into my formula (X~ treatment*time ; x~treatment/time ; or x~ treatment , strata = time)
My codes are:
>fisheatsfishnestfisheats2
PERMANOVA would be best because it tests the homogeneity of multivariate dispersions within groups on the basis of any resemblance measure and does not require normality of data. In your case, sites are the open or closed tanks imo :)
Hello Jiji,
It seems like you want to do a repeated-measure analysis. In this case I think you need to restrict the permutations that build your null distribution. The time points should only be permuted so that neighboring samples are 'swapped' during permutations.
Here is a nice blog post that explains repeated measurements PERMANOVA using adonis. There is some code included, but it's an old post so some things may be deprecated. Good luck!
http://thebiobucket.blogspot.com/2011/04/repeat-measure-adonis-lately-i-had-to.html
- Badges
- Science method
- Similar topics
- Biological Science
- Ecology