Skip to content
Prev 1306 / 7420 Next

some MDS and R problems

Arnaldo,

     Given that NMDS operates essentially on rank distances, small 
variations in dissimilarity generally have low impact.  I rarely have 
points with zero dissimilarity on quantitative indices, but it sometimes 
happens on presence/absence based indices.  If you have a dissimilarity 
object named dis, I just do

 > dis[dis==0] <- 0.0001
 > res <- nmds(dis)

Following Jari's argument about the other point's contribution to 
stress, I think fudging one dissimilarity is a better solution than 
dropping out a point

Dave Roberts
Arnaldo Russo wrote: