Social network packages poll

May 6, 2010 at 4:43 am 14 comments

| Gabriel |

Unfortunately (for now) Stata has basically no social network capabilities (though for some early efforts see Rense Corten’s netplot program and .net import filter, as well as my own .net export filter). So this necessitates leaving the warm cocoon of Stata and learning something else and a grad student recently asked me by email what social network package I recommend. Given that there’s a (steep) learning curve to all these alternatives, I think it’s reasonable to pick just one, at least to start with.

Personally, I use igraph running through R because it’s free, it’s easy to pipe to it, it can do alpha centrality, the documentation includes examples and tutorials rather than just a dictionary of functions and arguments, and it has a relatively easy syntax for importing Pajek data and generating graphs. You can see my attempts to muddle through this here. However I’m not going to claim that this is the best and it’s entirely possible I would have been better off grappling with something else (I especially have these “grass is always greener” thoughts about running igraph through Python). My only strong opinion is that whatever you use, it should be scriptable even if it comes at the cost of a higher initial learning curve.

Anyway, what do you all prefer and why? I’m allowing repeated voting for the benefit of people who use several. I’m interested in what people like in both a “wisdom of the crowd” sense and a network externalities sense.

About these ads

Entry filed under: Uncategorized. Tags: .

Misc links (niche partitioning and categories edition) Cited reference search time-series

14 Comments

  • 1. Rense  |  May 6, 2010 at 7:06 am

    Nice idea!
    I voted Pajek, because of its ability to handle large networks, and my inability to use R effectively (still learning, though). Apart from UCInet, I cannot judge the other packages because I never tried them seriously. Paradoxically, I absolutely agree with the “scriptable” argument, even though this is lacking in Pajek.

  • 2. jimiadams  |  May 6, 2010 at 9:04 am

    i still use SPAN, but mainly because it’s what i learned on (he being my advisor and all). However, i am slowly transitioning to the R-based packages – that general environment is increasingly useful (Siena now is available in R too, and they migrated Sonia to there as well – for those who do dynamic network stuff).

  • 3. Social Network Packages Poll « Permutations  |  May 6, 2010 at 5:19 pm

    [...] Social Network Packages Poll Gabriel Rossman is running it here. [...]

  • 4. Michael Bishop  |  May 6, 2010 at 5:36 pm

    Perhaps you should ask a second question, which package do you anticipate being most helpful in the coming years?

  • 5. Jesse  |  May 6, 2010 at 6:30 pm

    I’m a big user of R – igraph, but I’m currently warming up to ORA. It’s really great if you have multiple levels or kinds of relationships between the same set of nodes.

  • 6. Ted  |  May 9, 2010 at 7:57 pm

    I’ve started doing network analysis by using UCINET. It seems easy to get a handle on.

    I’m still a novice in SNA – do you think it’s a waste of time to continue working in UCINET?

    • 7. gabrielrossman  |  May 10, 2010 at 1:17 am

      ted,
      i’ve actually never used UCInet. anybody who has experience with both it and one of the newer packages want to take this?

  • 8. david schaefer  |  May 10, 2010 at 3:11 pm

    I use both R packages. I prefer the R environment itself not just because of the scripting but because it is versatile enough to do any other data management tasks a project requires.

  • 9. michal  |  May 12, 2010 at 5:07 am

    Two thumbs up for the comment from David. If you use R you are getting access to all the other wonderful things that come with R. Using specialized package, like Pajek, UCINET etc requires constant going back and forth between network software and some other general analytical software (sas, spss, what have you).

    I also use both R packages. Mainly ‘igraph’ (and that is what i voted for) because I find the network data manipulation much easier than in ‘network’. I believe it is also much more efficient computationally than ‘network’ (Gabor Csardi showed couple of comparisons on UseR 2009). On the other hand with ‘network’ you get access to ERGM and all the other statnet packages written by people at Stats Dept at University of Washington. So igraph/network is not an easy choice.

    I’m actually working on reliable routines that will convert igraph objects to network and vice versa. Perhaps an ideal situation would be to have a an umbrella interface to igraph, network, graph etc. very similar as DBI is an interface to RMySQL, RSQLite etc…..

  • [...] do you use? May 12, 2010 tags: analysis, pajek, poll, sas, stata, ucinet by Michal See a the poll here by Gabriel Rossman at Code and [...]

  • 11. Jeromy Anglim  |  May 13, 2010 at 3:50 am

    Is it possible to give three thumbs up?
    I agree with Michal and David.
    The scripting and general data analytic tools for R are excellent.
    To take one of my own examples, I had 8 classrooms at 3 time points with 4 different social network questions.
    R made it easy to write a script which displayed plots and network statistics for all 8 * 3 * 4 = 96 networks.
    If you want to do reproducible research, R is also a great option.

  • 12. Peter McMahan  |  May 30, 2010 at 9:18 pm

    Good poll, thanks!

    First, an appeal for another option: “R (sna)”. ‘sna’ is a really nice simple package if you are dealing with small networks, and can do cycle and triad census easily.

    Second, it was hard to pick just one of these. The right tool for the job really depends on the network question you’re trying to answer. I tend to use either igraph or sna depending on the scale of what I need done, and either python or R as the frontend for igraph depending on what I need to do with the results of the network analysis.

    • 13. gabrielrossman  |  May 30, 2010 at 11:28 pm

      i was implicitly including the sna package as part of the statnet metapackage — i can see how that would be confusing though.

  • 14. Assorted Links « Permutations  |  June 4, 2010 at 12:51 pm

    [...] back on Code and Culture’s Network Analysis Software Poll (igraph and statnet are the top [...]


The Culture Geeks

Recent Posts


Follow

Get every new post delivered to your Inbox.

Join 1,472 other followers

%d bloggers like this: