[R] Kruskal's MDS results

[Jari Oksanen]

Dieter Vanderelst <Dieter.Vanderelst <at> ua.ac.be> writes:

> 
> A few people suggested taking a look at Ripley's book MASS. I know the formula
listed there.
> 
> The point is that the manual for the isoMDS function says it's stress output
is in "percent". Does this mean,
> the stress reported by isoMDS is just the stress value in MASS (which ranges
from 0 to 1) value multiplied by
> 100? I've haven't been able to find any resource that expresses stress in
values from 0 to 100. So, this
> would be a convention introduced by the authors of the package?

Yes, it is stress multplied with 100 which makes it "per cent" instead of "per
unit". You can see this also in the software. The extract is from VR_mds_fn
function in MASS.c:


    sstar = 0.0;
    tstar = 0.0;
    for (i = 0; i < n; i++) {
        tmp = y[i] - yf[i];
        sstar += tmp * tmp;
        tstar += y[i] * y[i];
    }
    ssq = 100 * sqrt(sstar / tstar);
    
The last line has the multiplication, the previous collect the terms for the
stress. 

Best wishes, Jari Oksanen