Convex Hulls vs Gaussian Confidence Ellipsoids

[John Brazner]

This is more of an ordination question but I'm hoping someone might be able to provided guidance here. I've plotted convex hulls around different groups of site types in a NMDS ordination as a way of visualizing differences among the bird communities associated with these groups and used MRPP to examine the significance of differences among the centroids of those groups. I've been told by a colleague that using Gaussian confidence ellipsoids would be a better approach but It seems to me that the statistical rigour would be greater using MRPP given the resampling method used to obtain the statistical estimates and no assumptions about multivariate normality. Any thoughts would be appreciated.

John  

[Bruce McCune]

John, I completely agree with your assessment. Confidence ellipses assume that the points within a group are normally distributed in the ordination space, which is usually a poor assumption. There is often a gross mismatch between confidence ellipse overlays and the actual distribution of points in the ordination space. This is the reason that they are not offered in PC-ORD. Convex hulls are a conservative way of showing group overlap graphically and MRPP (or perMANOVA) as you suggest is a good way to assess statistical significance among groups.

-Bruce McCune

On Wed, Jun 24, 2020 at 12:34 PM John Brazner <braz...@gov.ns.ca> wrote:

This is more of an ordination question but I'm hoping someone might be able to provided guidance here. I've plotted convex hulls around different groups of site types in a NMDS ordination as a way of visualizing differences among the bird communities associated with these groups and used MRPP to examine the significance of differences among the centroids of those groups. I've been told by a colleague that using Gaussian confidence ellipsoids would be a better approach but It seems to me that the statistical rigour would be greater using MRPP given the resampling method used to obtain the statistical estimates and no assumptions about multivariate normality. Any thoughts would be appreciated.

John  

--
You received this message because you are subscribed to the Google Groups "HyperNiche and NPMR" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hyperniche+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/hyperniche/13e7bc1f-972c-4e76-aa6c-5437c65d6f72o%40googlegroups.com.