Suggested answers for
case study 3
|
The long and short forms seem to be placed within a continuum of
length/width ratios, and there does not seem to be any justification
for splitting the sample at any particular point.
Graphical comparison of data sets
The three samples are overlapping in the plots, but the Swedish
S. linnarssoni are usually larger than the S. glaber.
In addition, aspect ratios are perhaps differing a little between
the Norwegian S. linnarssoni and the S. glaber, at
least for small specimens.
Principal Components Analysis
98.8 percent of the variation in the sample is accounted for by
the first component. The other components are of minor significance.
Similar loadings of all variables on the first component means that as we go
along the principal axis, all measurements are increasing similarly. This
axis (component) is therefore a general size axis.
The loadings on the second component show that as we go along the
second axis, length is rapidly decreasing while the width values
are increasing. The second component can therefore be described as
a length/width ratio axis.
The scatter plot again shows some overlap between the groups, but
the S. glaber and the Norwegian S. linnarssoni are now
better separated than in the simple X/Y plots.
The 'typical long' (black dot 1) and 'typical short' (black dot 2) forms are
positioned far down and far up on the second component axis, in accordance
with our interpretation of this component.
Principal Coordinates Analysis
The groups are not really separated along the arch, and PCO does
not seem to work well in this case.
Cluster Analysis
For the Euclidean distance measure, overall size will be all important
for the clustering. As expected from our results above, the three
samples do not form well separated clusters.
Discriminant analysis
The discriminant analysis histogram does not show really good separation of
the two suggested species, but he Hotelling's t2 test reports a highly
significant difference.
Conclusion
The analysis from these data alone is somewhat inconclusive. There
is definitely a difference between the groups, but they are all
within a continuously overlapping range. PCA analysis is perhaps the
most revealing method for this data set.