NMDS analysis

Non-metric multidimensional scaling without writing code

NMDS is one of those analyses that has always meant R, a distance matrix and a script nobody else on the lab can read. Magic Stat runs it from the data you already have — you pick the dissimilarity measure, it gives you the stress value and the ordination figure.

The honest part

What NMDS actually is, and when it is the right tool

Non-metric multidimensional scaling takes a table of samples and variables and represents the samples as points in a low-dimensional space — usually two dimensions — so that the ranking of distances between points matches the ranking of dissimilarities in your data. It does not try to reproduce the distances themselves, only their order. That is what makes it "non-metric", and it is why NMDS copes with strongly non-linear, ecological, messy data where PCA distorts.

The number that matters is stress: how badly the two-dimensional picture distorts the original dissimilarities. Low stress means the picture is trustworthy; high stress means you are looking at a caricature. Magic Stat reports the stress alongside the figure, because an ordination figure without its stress is a picture pretending to be a result.

It is the standard tool in community ecology, microbiome work and any field where you have many species/variables measured on many samples and you want to see whether the groups separate. If your variables are continuous and roughly linear, PCA is usually the more honest answer — and Magic Stat does that too.

What it does

What Magic Stat gives you for NMDS

How it works

From a spreadsheet to an ordination figure

  1. Load your data. a normal spreadsheet — samples in rows, species or variables in columns. No reformatting, no transposing, no as.matrix().
  2. Choose the dissimilarity and the transformation. Bray–Curtis and Hellinger are the usual starting points for abundance data; the list is in the dialog.
  3. Run it. Magic Stat computes the distance matrix and the ordination in one step.
  4. Read the stress. the value is reported with the result, so you know whether the two-dimensional picture can be trusted.
  5. Export. the ordination figure goes to the gallery and the analysis goes into the report with its settings and interpretation.
Options

The settings, out in the open

Dissimilarity measures: Bray–Curtis · Euclidean · Manhattan · Canberra · Jaccard · Sorensen · Horn–Morisita.

Transformations: None · Z-score (standardise) · ln(1+x) · ln(x) · Hellinger · Chord · Chi-square · Square root · Range 0–1.

Frequently asked

NMDS questions, answered honestly

Do I need to know R?

No. That is the entire point. If you already use R comfortably and it serves you well, keep using it — Magic Stat competes on the analysis-to-manuscript workflow, not on replacing your scripts.

Which dissimilarity should I use?

For species abundance data, Bray–Curtis is the convention and the safest default. For presence/absence, Jaccard or Sorensen. For standardised, continuous variables, Euclidean. The measurement should match the data, not the other way round.

What stress value is acceptable?

Stress below 0.1 is generally considered good, up to about 0.2 is usable with care, and above that the configuration is distorting enough that it should be reported as a caveat rather than as a result. Magic Stat reports it so you can make that call yourself.

Can I use it for a paper?

The figure is exported at publication quality (including TIFF) and the analysis is written into a report with its settings, so the method section can be described accurately.

Is my data uploaded somewhere?

No. The analysis runs on your computer. The only signal the app sends automatically is an anonymous installation counter — no data, no names, no e-mail.

PCA analysis → PCoA analysis →