CODEWITHSUNDEEP
rvegan
outside842
outside842

Reputation: 51

Interpretation of ordiellipise NMDS

I am using metaNMDS to explore a multivariate dataset I am working with. I have constrained the dataset by a factor of interest that has 6 levels. The ordination is being run on a matrix of bray-curtis dissimilarities, is not being autotransformed, has 2 dimensions (or axises), and is set to run for a max of 300 iterations. The ellipses are based on 95% confidence and use the SE.

The NMDS is solved in about 10-15 iterations and has an okay to poor stress value (15-18). When I plot the data with ordiellipse to visualize which levels might be different I am struck by how little of the data are actually within the confidence ellipses. Can someone explain this? Is this just an artifact of the ordination is not providing a good fit to the data; not capturing the variation inherent in the dataset in 2-dimensions?

Any thoughts? My reputation isn't high enough to post a picture of the plot but I can send one out if I get a few bumps.

Upvotes: 2

Views: 2196

Answers (1)

Gavin Simpson
Gavin Simpson

Reputation: 174918

You are plotting a confidence ellipse for the mean (group centroid), which tells you something about the sampling distribution of the mean (centroid) you might see if you repeated your data collection a lot of times. In other words you are looking at the uncertainty in the estimate of the population mean (centroid) given the sample of data you collected.

The other type of confidence ellipse is based on the standard deviation. That is a measure of spread of the data (not the mean) and hence if you switch to that type your confidence ellipses should have less surprising coverage.

Note that this is the same misunderstanding the people have when discussing the standard deviation and the standard error of the mean in descriptive statistics.

Upvotes: 5

Related Questions