[cherdano]

This data won't be complete before we have seen mike777's scores

[helene_t]

[gwnn]

Helene beat me to it!

[Lobowolf]

Gandhi should be an honorary BBF regular.

[hrothgar]

helene_t, on Nov 17 2008, 11:44 PM, said:

Hey there

Thanks for creating the graph

Do me a favor and post the raw numbers (Yes, I'm lazy).
I'll see if I can find any interesting MATLAB functions to crunch the numbers

Maybe I can calculate some centroids or some such

[helene_t]

Hierarchical clustering, euclidian distances:

[helene_t]

hrothgar asked me to post this said:

I ran the raw data that Helene and Tim gathered through a cluster analysis program.

I arbitrarily told the program to divide the datapoints into four distinct clusters (This decision was quite arbitrary...  I based it on the fact that the political compass is based on two dimensions / four quadrants)

The four resulting clusters are color coded

The open blue circle next to MickyB is the centroid of the cluster 2
The open green circle slightly to the right is the centroid of all the participants

[matmat]

so we're all a bunch of liberal wieners.

[gwnn]

Can you do something with the hierarchical clustering to make it a little more legible?

[MickyB]

Yay, I aimed closest to the middle! Do I get a prize?

[matmat]

MickyB, on Nov 17 2008, 05:40 PM, said:

Yay, I aimed closest to the middle! Do I get a prize?

congrats. you're more average than anyone else!

[TimG]

helene_t, on Nov 17 2008, 05:01 PM, said:

hrothgar asked me to post this said:

I ran the raw data that Helene and Tim gathered through a cluster analysis program.

I arbitrarily told the program to divide the datapoints into four distinct clusters (This decision was quite arbitrary...  I based it on the fact that the political compass is based on two dimensions / four quadrants)

The four resulting clusters are color coded

The open blue circle next to MickyB is the centroid of the cluster 2
The open green circle slightly to the right is the centroid of all the participants

Looks like it would be more interesting to see this without Mbodell's point.

[Aberlour10]

OK, I surmounted my own laziness and did it too: -1,45, -3,08

It would be interesting to know which place would take "Maggie_T" on Helen's graphics

Robert

[y66]

Nice graphics!

[jonottawa]

-0.62/-4.26

Pretty centrist. First time I took it I was just right of center. Don't feel strongly about 5 or 6 of the questions. I think it's my ixnay on the limitedunnay ocreationpray for etardsray that cost me a higher libertarian score.

[helene_t]

gwnn, on Nov 17 2008, 11:07 PM, said:

Can you do something with the hierarchical clustering to make it a little more legible?

hmmm ... maybe it's possible to spread the clusters more apart.

Anyway, hierarchical clustering sucks. A mixture model (such as what I think Richard used) is more appropriate for this purpose.

I wondered if there was any clustering at all. Richard's results are one main cluster and a few outliers, which suggests there is no significant clustering.

On the other hand the statistic of Euclidean distances is bimodal, something that does not come up in simulated data. That is a rather ad-hoc test though. (In the meantime, Robert has arrived in the middle of the diagram and further dilutes the clustering).

[hrothgar]

helene_t, on Nov 18 2008, 01:31 PM, said:

gwnn, on Nov 17 2008, 11:07 PM, said:

Can you do something with the hierarchical clustering to make it a little more legible?

hmmm ... maybe it's possible to spread the clusters more apart.

Anyway, hierarchical clustering sucks. A mixture model (such as what I think Richard used) is more appropriate for this purpose.

I wondered if there was any clustering at all. Richard's results are one main cluster and a few outliers, which suggests there is no significant clustering.

On the other hand the statistic of Euclidean distances is bimodal, something that does not come up in simulated data. That is a rather ad-hoc test though. (In the meantime, Robert has arrived in the middle of the diagram and further dilutes the clustering).

Actually, I used a heirachical clustering... I could run kmeans, but there doesn't seem to be much point with so few data points.

The keu point is that there doesn't seem to be any (obvious) cluster to the data. It looks as if there is a fairly large diffuse cloud with some outliers.

One of these days I should give some thought to some kind of dynamic algorithm to select the number of clusters...

[helene_t]

hmmm ... your four clusters are different from my four top clusters. Somehow we must have used different parameters. Did you use some other metric than the Euclidean?

[hrothgar]

helene_t, on Nov 18 2008, 02:30 PM, said:

hmmm ... your four clusters are different from my four top clusters. Somehow we must have used different parameters. Did you use some other metric than the Euclidean?

I think that I used Euclidean distance

It's difficult to read dendograms and the results aren't always intuitive. You might want to rerun your analysis and specifically configure the system to find four clusters

[helene_t]

We could also have used different agglomeration method.

Anyway, let's just assume the two scores are beta-distributed and independent. The data set is too small to support anything more complex.