Different Centralities Results from Les Miserables between version 113 and version 226

Mar 14, 2013 at 8:15 PM
Hello, I'm self-learning the version 226 to teach an undergrad course in Communication Studies and have a question regarding the different results of centrality scores between the version113 (that the textbook is based off) and this version (226).

The textbook figure 5.6 shows that some of the centrality results and the scores are different from what I got from 226 version (see here hyperlinke; The result was computed based on undirected graph). In particular, while clustering coefficient and eigenvector centrality are slightly different, betweenness is far different (Is it because the previous version is based on normalized value and the 226 is not?).

Given the rank-orders of the results seem consistent, I'm speculating that there might have been some changes in formula for computation? I would like to have some simple answer for the differences so that I can explain students about the differences and avoid confusion. (The students are using the textbook as a reference).
Mar 14, 2013 at 8:39 PM

You've guessed correctly. Since version 113 was published, which was more than three years ago, we've switched to using a high-performance software package called "SNAP" (http://snap.stanford.edu/) to calculate some of our graph metrics. In some cases, the package uses different formulas from what we were using in 113. For betweenness centrality, in particular, I believe version 113 was normalizing values, and newer versions of NodeXL don't do that.

-- Tony
Mar 14, 2013 at 9:43 PM
Thanks Tony, much appreciated! -Hazel