by Frau_Blucher » Mon Oct 16, 2006 3:26 am
[[b]MASSIVE EDIT!!!</B> Had some misassumptions. Please see new chart on page 2]
Here's the big daddy - The Cluster. First, several caveats. Heirarchical clustering tries to bin things that go together. In this case, we're going to bin YOU, the submittors to the poll. Clustering is also extremely qualitative and is as much a measure of things that are different as they are things that are similar. This is why things that look [i]apparently[/i] similar often don't get placed in the same bins. Capisci?
Anyway, I used a Rank Order clustering that is good for items that contain non-natural numbers, like things that are ranked as here. I ran several other types of clusterings and found they generally gave very similar results. The correlation percentages are also qualitative on a scale of Plus-100% to Minus-100% and means nothing in the absolute. The closer you are linked to someone near the bottom of the Tree (or Dendogram), the more similar you are. Being next to somone a Branch over is not necessarily an indication of similarity - you have to go back up your Branch and see where their Branch links to your branch.
The color intensities in the matrix signify Points in this case - the more intense the Red, the greater the points that that Song/Band got. Blacks have very few points. Grays are null values. I've also computed the Average of points that any Song/Band got and made that into an [i]individual[/i] so you can see your comparison to the Average. Oh, about Bands - I decided to include them as an individual metric in addition to the Songs with the rationale that with so many different songs, you're likely to pick the same band but not the song. That really shouldn't yield zero similarity since you did at least pick the same Band. I've therefore given a partial weighting to the Band of 1/3 of the points that that Band's song got. Wit me?
There's a bazillion patterns here - I'd love to hear others' interpretations of them as well. This is really the fun part - for everyone else to discuss eachothers' observations. GO!
[[b]MASSIVE EDIT!!!</B> Had some misassumptions. Please see new chart on page 2]
<img src="x">