Multidimensional Scaling

I've 5x14 data matrix. I'm using the MDS to get a perceptual map. I can do the MDS properly & get the result.

But my problem is in MDS we can map either row or column variables. Is it possible to map both row & column variable using MDS.

The code I used is the following:

perp<-read.csv("E:\\Projects\\Combined_3.csv")
ads.dis<-dist(perp)
perp_mds <- cmdscale(ads.dis, 开发者_如何学JAVAk = 2,eig=TRUE)
x <- perp_mds$points[,1]
y <- perp_mds$points[,2]
plot(x,y, xlab = "Coordinate 1", ylab = "Coordinate 2", type = "n")
text(x,y, labels = rownames(perp))

I'll be grateful if somebody can help me with the coding.

Regards, Ari

In general, the answer is no, not with cmdscale(). All that cmdscale() has knowledge of is the dissimilarity between objects. In the vegan package, there is function capscale() which is a constrained version of principal coordinates analysis (PCoA aka MDS), but can be used for normal PCoA. It can place both the objects and the variables in a biplot-like figure:

require(vegan)
data(varespec)
mod <- capscale(varespec ~ 1)
plot(mod)

But do note that PCoA with the euclidean distance is the same as PCA, which also could be used and will naturally plot both the objects and the variables:

plot(rda(varespec))

or using base R functions

mod2 <- prcomp(varespec)
biplot(mod2)

Or did you mean the non-metric version of MDS?