Table 6. Based on these means, it appears that for each of the five approaches,
the clustering algorithm finds groups of high, medium and low achieving children.
Interestingly, however, the degree of group separation is not uniform across methods.
For example, it appears that the Jaccard, Russell/Rao and Dice methods
all result in three clearly defined groups based on the test scores. On the other
hand, the Matching coefficient cannot seem to distinguish between the highest
achieving group, and one that achieves in the middle. It does, however, clearly
group the lowest achieving students. Finally, the raw data approach does appear
to successfully differentiate the groups based on the reading and math scores, but
not general knowledge.
Table 6: Mean scores on achievement tests by distance measure and
cluster type