Graphics Reference
In-Depth Information
in Cox and Cox (
), Borg and Groenen (
) and elsewhere. A configuration is
found that minimizes S, usually using a gradient descent approach.
he rail data used for classical scaling were analysed using non-metric MDS. Fig-
ure
.
shows the configuration obtained, and again the solution is arbitrary up to
translation, rotation and reflection. he STRESS associated with the optimum so-
lution is
%. It should be noted that
randomly selected starting points were
employed to ensure that the true optimum has been obtained.
AShepardplotmayalsobeutilized toassesstheprocedure.hisissimplyaplotof
d
rs
and d
rs
against δ
rs
and is shown in Fig.
.
for the rail station data. It shows how
well the distances within the configuration match the original dissimilarities accord-
ing to rank order. It makes the monotonic least-squares regression fit particularly
clear by joining the δ
rs
and d
rs
pairs.
he next section gives a more detailed example and shows how the quality of the
fit of the model can be investigated.
Figure
.
.
A map of rail stations from non-metric MDS