On clustering ‘Try Books!’ people

Agglomerative clustering: better with left-click

We can also look at grouping people based on the ‘distances’ between their scores.  The distances are going to be a mixture of how people felt about the books and what their scoring conventions are:  if A and B both use 5 to mean ‘not very good’, 7 to mean ‘OK’ and 8 to mean good, then the distances are going to mean something real, while if B indicated ‘not very good’ by 3, ‘OK’ by 5 and 9 to mean ‘good’, then a lot of the distance between them will be down to  different conventions.  We are also confine to a subset of members such that everyone shares at least one scored book with each of the other members (or we’ll end up dividing by zero).  The subset presented here isn’t the only one possible.

Anyway, the diagram above shows  the results of agglomerative clustering, while the one below shows those obtained from divisive clustering.

Divisive clustering: left-click advised

These two look reasonably similar, which is reassuring.  There are two stable lower-level groups in Candida/Aruni/Jo/Rob and Judy/Jane.  It’s not clear what the higher-level structure here means, but then the original ‘distances’ are not so easy to interpret either.  Apart from Judy/Jane, we don’t see that much association between the people who in theory come together, nor is there much evidence of solidarity amongst the token males.

Finally, we can consider multi-dimensional scaling:


This also displays the grouping Candida/Aruni/Jo/Rob, who must win a prize for being consistent agreement under a variety of analyses.  Otherwise there’s not a great deal of easily-interpretable structure!


Tags: ,

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: