Udell exploration

Jon

I used your data to whack out a square matrix^[1] and , ran it through a social network analysis program called ucinet, and made some pictures.

Two dimensional multi-dimensional scaling gives you this.

Three-D factor analysis yielded this table, which I visualized using VPython ^[2]

This looks much better if you run the program and animate the display (Rt-Mouse drag on the screen to rotate). At that point (only) the natural groupings seem to stand out.

But are they natural groupings? You know these people?

A heirarchical cluster analysis is interesting may be the best and simplest visualization of all, since the “only reality” is the similarity of nodes to nodes and groups of nodes.

Ucinet details follow:

METRIC MULTIDIMENSIONAL SCALING

Picture

Data

--------------------------------------------------------------------------------

Starting config: GOWER'S PRINCIPAL COORDINATES

Type of Data: Similarities

Input dataset: C:\Program Files\Ucinet 5\DataFiles\udellsim

15 items

Initial Stress = 0.894

Final Stress = 0.266 after 6 iterations.

1 2

------ ------

1 Thierry Lalinne 0.398 0.550

2 Paul Snively 0.308 0.059

3 CE Granier 0.055 -0.068

4 David Brown 0.117 0.032

5 Joe Jennet 0.082 0.546

6 Jim McGee 0.107 0.329

7 Jenny Levine -0.049 0.095

8 Sam Ruby 0.635 0.270

9 Jiri Ludvik 0.425 -0.175

10 Jon's Radio 0.383 0.441

11 Olivier Travers -0.082 0.294

12 Gordon Weakliem 0.629 0.200

13 Peter Drayton 0.526 0.207

14 Dann Sheridan 0.157 -0.189

15 Marc Barrot 0.569 -0.058

Coordinates saved as dataset MetricMdsCoord

----------------------------------------

Running time: 00:00:01

Output generated: 30 May 02 16:55:34

JOHNSON'S HIERARCHICAL CLUSTERING

Data

--------------------------------------------------------------------------------

Input dataset: C:\PROGRAM FILES\UCINET 5\DATAFILES\udellsim

Method: AVERAGE

Type of Data: Similarities

HIERARCHICAL CLUSTERING

O T G

l h o

i D i r P

v P J a e d e

i a D e n r J o t J M

J e u C a n n r o n e i a

o r l E v J n y n r r r

e i i y S ' W S i c

T S G d m h L s e a D

J r n r L e a a m r L B

e a i a B M e r l R k a u a

n v v n r c v i i a l R y d r

n e e i o G i d n d i u t v r

e r l e w e n a n i e b o i o

t s y r n e e n e o m y n k t

1 1 1 1 1 1

Level 5 1 2 3 4 6 7 4 1 0 2 8 3 9 5

------ - - - - - - - - - - - - - - -

46.154 . . . . . . . . . . XXX . . .

36.551 . . . . . . . . . . XXXXX . .

33.333 . . . XXX . . . . . XXXXX . .

22.437 . . XXXXX . . . . . XXXXX . .

21.858 . . XXXXX XXX . . . XXXXX . .

20.000 . . XXXXX XXX . . . XXXXX XXX

19.524 . . XXXXXXXXX . . . XXXXX XXX

18.182 . . XXXXXXXXX . XXX XXXXX XXX

17.449 . . XXXXXXXXXXX XXX XXXXX XXX

14.121 . . XXXXXXXXXXX XXX XXXXXXXXX

12.173 . XXXXXXXXXXXXX XXX XXXXXXXXX

11.405 . XXXXXXXXXXXXX XXXXXXXXXXXXX

8.730 . XXXXXXXXXXXXXXXXXXXXXXXXXXX

6.017 XXXXXXXXXXXXXXXXXXXXXXXXXXXXX

Clustering permutation saved as dataset hicluspermutation

5 11 2 3 4 6 7 14 1 10 12 8 13 9 15

Joe Jen Olivier Paul Sn CE Gran David B Jim McG Jenny L Dann Sh Thierry Jon's R Gordon Sam Rub Peter D Jiri Lu Marc Ba

------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- -------

5 Joe Jennet 100.000 5.578 4.348 8.791 10.501 15.789 10.314 8.617 15.000 8.252 3.731 5.479 4.826 2.857 6.017

11 Olivier Travers 5.578 100.000 12.552 10.924 12.017 16.722 15.969 12.173 6.167 11.439 5.808 6.364 9.382 7.373 7.424

2 Paul Snively 4.348 12.552 100.000 20.253 22.437 15.714 21.138 15.873 17.647 9.009 12.361 16.393 16.194 17.241 15.385

3 CE Granier 8.791 10.924 20.253 100.000 33.333 15.827 21.311 25.806 8.955 3.636 0.000 0.000 7.579 14.035 7.843

4 David Brown 10.501 12.017 22.437 33.333 100.000 15.472 19.524 19.243 12.130 7.472 6.253 6.061 9.570 14.168 13.272

6 Jim McGee 15.789 16.722 15.714 15.827 15.472 100.000 21.858 11.382 10.938 18.713 5.594 6.612 9.161 6.780 5.357

7 Jenny Levine 10.314 15.969 21.138 21.311 19.524 21.858 100.000 17.449 9.830 9.187 2.602 3.846 7.370 6.220 8.771

14 Dann Sheridan 8.617 12.173 15.873 25.806 19.243 11.382 17.449 100.000 3.922 8.415 4.767 4.545 6.047 14.634 8.730

1 Thierry Lalinne 15.000 6.167 17.647 8.955 12.130 10.938 9.830 3.922 100.000 18.182 9.879 12.245 8.237 8.696 9.565

10 Jon's Radio 8.252 11.439 9.009 3.636 7.472 18.713 9.187 8.415 18.182 100.000 11.887 8.696 12.845 8.989 11.405

12 Gordon Weakliem 3.731 5.808 12.361 0.000 6.253 5.594 2.602 4.767 9.879 11.887 100.000 46.154 36.551 10.826 17.374

8 Sam Ruby 5.479 6.364 16.393 0.000 6.061 6.612 3.846 4.545 12.245 8.696 46.154 100.000 28.571 10.256 12.121

13 Peter Drayton 4.826 9.382 16.194 7.579 9.570 9.161 7.370 6.047 8.237 12.845 36.551 28.571 100.000 10.817 14.121

9 Jiri Ludvik 2.857 7.373 17.241 14.035 14.168 6.780 6.220 14.634 8.696 8.989 10.826 10.256 10.817 100.000 20.000

15 Marc Barrot 6.017 7.424 15.385 7.843 13.272 5.357 8.771 8.730 9.565 11.405 17.374 12.121 14.121 20.000 100.000

Partition-by-actor indicator matrix saved as dataset Part

----------------------------------------

Running time: 00:00:01

Output generated: 30 May 02 17:03:52

Picture

K-CLUSTERS VIA TABU SEARCH

Data

--------------------------------------------------------------------------------

Diagonal valid? NO

Number of clusters: 2

Type of data: Proximities

Method: correlation

Input dataset: C:\Program Files\Ucinet 5\DataFiles\udellsim

Starting fit: 0.819

Starting fit: 0.484

Fit: 0.484

Fit: 0.484 (smaller values indicate better fit.

r-square = 0.266

Clusters:

1: Paul Snively CE Granier David Brown Joe Jennet Jim McGee Jenny Levine Olivier Travers Dann Sheridan

2: Thierry Lalinne Sam Ruby Jiri Ludvik Jon's Radio Gordon Weakliem Peter Drayton Marc Barrot

Clustered Data Matrix

4 2 3 7 5 6 11 14 1 9 10 12 13 8 15

David Paul S CE Gra Jenny Joe Je Jim Mc Olivie Dann S Thierr Jiri L Jon's Gordon Peter Sam Ru Marc B

------------------------------------------------------------------------------------------------------------

4 David Brown | 46.154 23.529 33.333 21.875 14.433 15.172 12.295 17.647 | 10.959 12.698 8.621 6.349 7.921 6.061 14.035 |

2 Paul Snively | 23.529 46.154 20.253 21.138 4.348 15.714 12.552 15.873 | 17.647 17.241 9.009 10.345 18.750 16.393 15.385 |

3 CE Granier | 33.333 20.253 46.154 21.311 8.791 15.827 10.924 25.806 | 8.955 14.035 3.636 12.632 7.843 |

7 Jenny Levine | 21.875 21.138 21.311 46.154 7.407 21.858 19.149 18.868 | 7.207 5.941 5.195 1.980 5.755 3.846 6.316 |

5 Joe Jennet | 14.433 4.348 8.791 7.407 46.154 15.789 5.578 8.000 | 15.000 2.857 4.878 2.857 5.556 5.479 3.125 |

6 Jim McGee | 15.172 15.714 15.827 21.858 15.789 46.154 16.722 11.382 | 10.938 6.780 18.713 5.085 11.538 6.612 5.357 |

11 Olivier Travers | 12.295 12.552 10.924 19.149 5.578 16.722 46.154 9.009 | 6.167 7.373 14.074 5.530 11.765 6.364 5.687 |

14 Dann Sheridan | 17.647 15.873 25.806 18.868 8.000 11.382 9.009 46.154 | 3.922 14.634 8.511 4.878 5.063 4.545 11.429 |

--------------------------------------------------------------------------------------------------------------

1 Thierry Lalinne | 10.959 17.647 8.955 7.207 15.000 10.938 6.167 3.922 | 46.154 8.696 18.182 8.696 7.143 12.245 10.000 |

9 Jiri Ludvik | 12.698 17.241 14.035 5.941 2.857 6.780 7.373 14.634 | 8.696 46.154 8.989 11.111 10.811 10.256 20.000 |

10 Jon's Radio | 8.621 9.009 3.636 5.195 4.878 18.713 14.074 8.511 | 18.182 8.989 46.154 13.483 17.323 8.696 12.048 |

12 Gordon Weakliem | 6.349 10.345 1.980 2.857 5.085 5.530 4.878 | 8.696 11.111 13.483 46.154 40.541 46.154 20.000 |

13 Peter Drayton | 7.921 18.750 12.632 5.755 5.556 11.538 11.765 5.063 | 7.143 10.811 17.323 40.541 46.154 28.571 14.706 |

8 Sam Ruby | 6.061 16.393 3.846 5.479 6.612 6.364 4.545 | 12.245 10.256 8.696 46.154 28.571 46.154 12.121 |

15 Marc Barrot | 14.035 15.385 7.843 6.316 3.125 5.357 5.687 11.429 | 10.000 20.000 12.048 20.000 14.706 12.121 46.154 |

-------------------------------------------------------------------------------------------------------------

Partition saved as dataset TabuCluster

----------------------------------------

Running time: 00:00:01

Output generated: 30 May 02 17:10:28

^[1] print '---------'

print ' '.rjust(20),

for k1 in d.keys(): #colheads

(k1Url, k1Name) = k1.split(',')

print k1Name.rjust(20),

for k1 in d.keys():

(k1Url, k1Name) = k1.split(',')

print k1Name.rjust(20), #row head

for k2 in d.keys():

(k2Url, k2Name) = k2.split(',')

s.set_seqs(d[k1],d[k2])

print `s.ratio()*100`.rjust(20),

^[2] People=[

("Thierry Lalinne", -0.57, 0.24, -0.18),

("Paul Snively", 0.42, 0.02, -0.07),

("CE Granier", 0.56, 0.44, 0.05),

("David Brown", 0.44, 0.41, 0.00),

("Joe Jennet", -0.45, 0.38, 0.11),

("Jim McGee", -0.11, 0.20, 0.58),

("Jenny Levine", 0.44, 0.28, 0.48),

("Sam Ruby", -0.09, -0.78, -0.14),

("Jiri Ludvik" , 0.17, 0.09, -0.60),

("Jon's Radio", -0.49, -0.07, 0.12),

("Olivier Travers", 0.07, -0.02, 0.57),

("Gordon Weakliem", -0.08, -0.86, -0.19),

("Peter Drayton", 0.04, -0.76, 0.01),

("Dann Sheridan", 0.44, 0.34, -0.08),

("Marc Barrot", 0.04, -0.11, -0.60)

]

from visual import *

for p in People:

(name,x,y,z)=p

label(pos=(x,y,z),text=name,height=14,box=0,opacity=0,color=(0,1,1))

sphere(pos=(x,y,z),label=name,radius=0.05)

Udell exploration

Jon

I used your data to whack out a square matrix[1] and , ran it through a social network analysis program called ucinet, and made some pictures.

Two dimensional multi-dimensional scaling gives you this.

Three-D factor analysis yielded this table, which I visualized using VPython [2]

METRIC MULTIDIMENSIONAL SCALING

Picture

Data

JOHNSON'S HIERARCHICAL CLUSTERING

Data

Picture

K-CLUSTERS VIA TABU SEARCH

Data

I used your data to whack out a square matrix^[1] and , ran it through a social network analysis program called ucinet, and made some pictures.

Three-D factor analysis yielded this table, which I visualized using VPython ^[2]