What package and steps needed to emulate legacy NTSYS ?
That sounds simply like a job for the base hclust() function. The help for that function has a worked example. But Euclidean distance is probably the wrong choice for mixed binary and continuous variables. I'd suggest looking into other options, like maybe Gower, and also rethinking whether both types of variables should be in the same analysis. Sarah
On Sun, Dec 14, 2014 at 6:52 AM, Bruce Miller <batsncats at gmail.com> wrote:
Hi all, Especially Jari.... There are now a lot of clustering algorithm packages in R. What would be the most usable package to run similar analyses as the older NTSYSpc? I was able to dig through an older backup image of one of my hard drives and resurrected NTSYSpc v. 2.11 but would prefer to set up similar analyses in R. To have similar results the starting matrix data frame would be OTUs (critters) as rows and the columns are values of things needing to be compared. These are comprised of both 1/0 data for locations and continuous variables for measurement data (critter sizes and in this case echolocation call parameters). Running NTSYSpc the initial data matrix may or may not first be standardized. Then a similarity coefficients are computed to generate a new matrix which is used as input for a clustering SAHN (UPGMA using Euclidean distance) then graphically the tree is displayed (GGPLOT2?) I am sure there is a way to do this, but have not found vignettes for the various R packages to explain code to set this up. Thanks for any suggestions. Bruce
Sarah Goslee http://www.functionaldiversity.org