NMDS: minimizing power or plotting multiple dimensions/axes

[hsuze22]

Hi all,

I'm a graduate student, working with a dataset "prim" of fish samples from four communities: After Treatment (AT), Before Treatment (BT), After Control (AC), and Before Control (BC). I have sites as rows and species, community ID, and 4 other factors as columns. I also created a metaMDS-ready dataframe "All_df" that has been narrowed to only include species abundance data.

I have 518 sites (rows) total and 40 species (columns) in All_df. Both "prim" and "All_df" have already had rare species (present at < 5% sites) removed.

PROBLEM:
I would like to perform NMDS and plot it. When I run the code with all four communities,:

It runs the 100 tries and I get the following message with stress values around 0.191:

*** Best solution was not repeated -- monoMDS stopping criteria:
74: no. of iterations >= maxit
26: stress ratio > sratmax

The only way I've been able to get best solutions repeated is by increasing k from 2 to 3 AND by subsetting the dataframe by communities and running each analysis with only 2-3 of the 4 total communities (e.g. Treatment plot with AT & BT combined; AT, BT, and AC combined).

I retried metaMDS with the many combinations of commids, and they all found repeated best solutions except the dataframe including all 4. I've provided an image below that shows the resulting stress values for some of those subset groups. I did this for combinations of two commids as well. In all cases, the only way I could get repeated best solutions was with k = 3 and excluding at least one commid.

I have two primary questions:

1. Are there ways to decrease stress and increase chances of repeated best solutions? I would like to plot all four commids together and/or acheive a k = 2 for plotting purposes (relates to my second question), but I currently get very high stress values and no repeated best solutions with all four commids as shown above and with subset data when k=2.

2. With k = 3, what is the best way to plot an nmds result? I prefer ggplot2, but cannot find a way to include a third axis. Is it innaccurate to plot an NMDS with k = 3 on a 2D graph?

Example: I plotted "Treatment Community Before and After" with plot_ly by combining my treatment commids c("AT", "BT") but I don't like having three axes or how it looks:

I've attached the plot below. I don't particularly like how it looks, and I'd really like to have a 2D plot, preferably with four commids on a single plot. Can I use ggplot2 to plot the first two dimensions and ignore the third?

Thanks, any guidance or perspective on the importance of preserving dimensions or ways to reach convergence with all 4 commids would be very appreciated!

[gaborcsardi]

Why are you posting this in the r-hub/rhub repository?

[hsuze22]

I am new to github, where should I post instead?

[gaborcsardi]

This discussion board is about the rhub R package, i.e. https://github.com/r-hub/rhub
If your question is related to a repository on GitHub, then post in that repository's discussion or issues. It certainly does not seem to be related to rhub.