NMDS is a non-parametric (no eigenvalues!) dimension reduction method that simply seeks to produce a plot where similar samples are close to each other and dissimilar samples are farther apart. Rather than preserving exact distances among samples, it preserves the ordered (rank) relationships. This method is applied on a distance matrix, which can be computed with any distance measure. It can work with variables that are numerical, ordinal or qualitative (or a mix of these). NMDS is a flexible method which avoids assumptions of linearity or normality, and is therefore a method of choice for skewed data, or zero-heavy data, such as commonly found in species abundance samples.
This workshop will focus on the application and interpretation of NMDS, with demonstrations in R on several example data sets.