evan_tech

Previous Entry Share Next Entry
08:11 pm, 24 Dec 04

bias

Don't trust my previous friendrank numbers. I discovered a bad bias in the way I compute it.


The LJ friends exports have both outgoing and incoming links. However, for a user on the fringe (like avva in the previous posts), I don't have data for all of the people he's connected to. So if I import all of the data, it looks as if each one of avva's friendofs has only one outgoing link--the one to him.

If I only use the outgoing links in the LJ friends data, I get an entirely different picture.
Unfortunately, that looks really biased as well; all the top scorers are my LJfriends. Here's the top 25 again:
                    user        rank
1                 tsenft        2.864640
2               sillyape        1.837850
3                   brad        1.344791
4                   evan        1.289011
5               holyloki        1.235474
6                patrick        1.127457
7                 mobley        1.080410
8               elliptic        1.064162
9             sleepsucks        1.039973
10              whitaker        1.004112
11                 faith        0.951336
12                  four        0.907252
13                 jeffr        0.896390
14               klayzen        0.856345
15              hellarad        0.788394
16        goldfischegirl        0.777827
17                   jag        0.760783
18             nefarious        0.751266
19            postmortal        0.723877
20                  lisa        0.707124
21                 meenk        0.688035
22          atypical_gnu        0.659139
23                  yiyi        0.645266
24            wraincloud        0.609160
25                 chris        0.589686

Needs more thinking.