I analyze the microbial community composition from soil samples collected in an agricultural field experiment. Within this field, there are 3 plots (randomly distributed in the field) and for each plot, we collected soil samples at 2 depths and 3 different compartments (comp). We have a total of 18 samples. BUT I only have one observation per sample unit. So, basically the replicates are the plots. My first idea was to write the following model with adonis:
x ~ depth + comp
With plot factor considered as replicate. But I am not sure it's right because the plots are distant (several meters). They are not true replicate, right? Then, after I had discussed with some colleagues, it turned out that I should maybe use the following model:
x ~ plot/depth + plot/comp + plot
depth and comp being nested in plot. But in that case, I have no replicate at the lowest level, and it might be a problem for a permutation test... I don't know how to solve this problem. The experimental design is a bit tricky because we have no replicates at the lowest level. Therefore, I don't know how to deal with the statistic model.
If anyone can help me I would greatly appreciate :) Thanks for your time!
Hi Florine,
I will give you my two cents on this although I am not 100% sure.
In my opinion, you can consider the 3 plots as replicates (biological not technical ones). Think about your experimental design: why have you decided to analyse 3 plots in the first place?
If you are nesting depth and compartment within plots you are basically saying that depth 1 of plot 1 is not equivalent to depth 1 of plot 2. For me a nested design would only make sense if would have sampled for example depths 1cm and 2cm in plot 1 and 5cm and 10cm in plot 2. If you do have the same depths for all plots than you have a traditional crossed design, not a nested one.
Sorry if that doesn't solve your problem.
In any case I am curious to see what others have to say about this as well.
Good luck!
Just to clarify on Igor's response; I think your advisors were probably trying to get at repeated measures rather than a nested design per se. Repeated measures is structurally the same as applying a blocking factor to the experimental unit (field). Nested design is similar to blocking, except that with a nested design your blocking factor is a treatment, and so you are interested in it's effect. Unless you are interested in the effect of your field choice, then I agree with Igor, it doesn't sound like your experiment is nested; it does however sound like it has repeated measures. Here's an example in R: https://www.r-bloggers.com/how-to-do-repeated-measures-anovas-in-r/
Dear Florine,
I attached the PERMANOVA manual which includes solutions and examples to how to deal with designs lacking replicates at the lowest level. In addition, from my personal experience, you can also directly send email to Prof. Anderson to make sure of your design and approach. I hope that helps.
Cheers!
- Badges
- Science method
- Similar topics
- Biological Science
- Ecology