Network Analysis, which command to use

[Charlie Joyez]

Hello everybody.
I have seen that an integrated network analysis tool was a popular request for Stata14, so maybe some of you know better than me what are the current methods of network analysis.

I've found several user-written network analysis tools :
- netplot by R. Corten (see article in Stata Journal here : http://www.stata-journal.com/sjpdf.h...iclenum=gr0048 ), which seems convienient to draw network graphs from edges list, but does not compute any statistics (on vertices centrality degree, closeness centrality, etc..). It doesn't seems to allow weights on edges, nor color on the graphs.

- netsis (and related netsummarize) (also an article in SJ : http://www.stata-journal.com/sjpdf.html?articlenum=st0248 ), which gives the above mentioned statistics, either through a mata or stata implementation. However, the graphs on the articles are made via netplot, so no graphics here.

-nwcommands by Thomas Grund, which is still not mentioned on Statalist, and seems to do both graphs and statistics description, but the programming seems very special, with networks being entities apart, beside variables. Here is the website of the project (said "network analysis using Stata") : https://nwcommands.wordpress.com/ . The graphics seems to be even more complete than with netplot (possible to use colors and weighted edges).

Is anyone familiar with these tools? The last one in particular? Do you recommand to use netsis and netplot for respectively statistical analysis and graphical representation, or should I better learn the nwcommands specificities and use it for both analysis.

Ps : I know it certainly depends on what I'm aiming to do. So, basically I want to describe FDI network (host country being the vertices, linked if the same firm invested in both), and to see how the centrality, etc.. of this network changes with the considered firms (as I'd like to highlight firm heterogeneity in FDI strategies).

Edit : I forgot to mention the networkplot command from Chaimani et al (2013) (http://journals.plos.org/plosone/art...l.pone.0076654), which seems to do great graphs. Although Network Analysis is not much used in Stata, there is many independently developed command, so I'm a little confused on wich one to use.. Any advice?

[Mike Lacy]

I've been experimenting with -nwcommands-,and it's my impression that it is vastly better than anything previously available in Stata. It is a whole system for network analysis, and implements a large number of standard network procedures. From the purely computation point of view, I'm amazed at how fast it does very hard calculations, such as centrality measures involving all possible paths between nodes. It has simple and fast import capabilities to accept network data in a large variety of formats. And, it has a user friendly set of dialog boxes.. I have not completely dug into the guts of it yet, but my experience is that networks are not the special entities that you experience, that is, one can easily move back and forth between having data in as a Stata data set, and having the relevant parts of that data in a network representation. What's going on is that networks are stored as Mata matrices, with Stata commands to manipulate them, following a Stata-ish syntax. I think it's a completely amazing tour de force, and that it will get rave reviews as it gets discovered. But, and this is a large but: I am a novice regarding networks.

Regards, Mike

[Charlie Joyez]

Thanks Mike, I really appreciate your very enthusiasm comment!
It what I thought, -nwcommands- seems to be very complete and it's great news that it is also fast.

I'll try to learn the its specificities. Do you have any other introduction material than the slides available on the nwcommands site?

Thanks anyway,

Best,
Charlie

[Thomas Grund]

Great that Mike (and hopefully others as well) seem to like the nwcommands. They are written in such a way that your normal Stata intuition should work. Hence, you deal with networks just as if they were variables (you can keep them, drop them, replace them...). You can create such network quasi-objects from edgelists, adjacency matrices and many other network file formats. Some additional features are that you can combine networks with everything else you normally do in Stata. Plotting networks (including node size, node color, node symbol, edge width, edge color...) and making movies of evolving networks is very easy as well. Furthermore, almost all nwcommands have their own dialog box as well. Hence, you can click your way through like in Ucinet if you want (but also go ahead and program with networks like in R).

The best resource is http://nwcommands.org. Check it out. There are some examples and further links.

I also uploaded several videos to Youtube (http://bit.ly/1Oj81GC). You can subscribe to the channel, where I regularly post video tutorials. This should give you a good start. Other than that I am currently writing up some tutorials that I can hopefully distribute soon.

There is also the nwcommands email list (http://bit.ly/1J7xSzD). This is a great list to ask questions concerning the nwcommands and to stay informed about developments and additional features.

Lastly, there are several full-day workshops coming up. The next one will be on 23 June at the XXXV Sunbelt conference in Brighton. You can register here: http://bit.ly/1HbFzpA. After that there will be a two-day workshop at the Stata Nordic summer school in Stockholm on 13-14 August (http://bit.ly/1FXCaJA). And there will be another one-day workshop at the Italian Stata User Group meeting in Florence on 13 November (http://bit.ly/1D8Drcy).

There is also a StataPress book in the making…, but more on that when it is ready.

Best,
Thomas

http://www.grund.co.uk

[Steve Vaisey]

This is fantastic news. Thanks, Thomas!

[Charlie Joyez]

Thank you Thomas for your answer, and welcome to Statalist.
I installed the -nwcommand- and the very first steps are indeed quite familiar you are right.

Some questions though. When creating a network from an edgelist (nwfromedge command), I observe that the data is reshaped in an adjacency matrix, to set the network.
but, third variables (besides the nwfromedge varlist) are dropped.
However, when I look at some pre-built network dataset. (e.g. webnwuse gang) , I don't see any adjacency matrices (although two networks are declared), and other variables besides the networks (Age, prison, music, etc..). I guess the matrices are in some mata format and not accessible through Stata browser.

So m question is, when I have my initial dataset (with an edgelist in it) and I want to declare a network but keep some other variables (that would be used for edgesize, (node)size options in the nwplot command.), how do I save the network created with nwfromedgelist, and import it in my initial dataset?

Thanks anyway, I'll keep reading and watching the videos.
The first results I got are great!

Best,
Charlie

[Charlie Joyez]

All right, I managed to deal with it, using the dialogue boxes, a very convenient tool to begin, before being familiar with code lines.
Let's consider this post closed, but I hope to read more on network analysis in general and nwcommands in particular here!
Thanks again Thomas for the great tool you provide!

Best,
Charlie

[Margarita Cabra]

Hi Charlie,
may I ask you how did you deal with the problem? I have the same issue now and don't know what to do.
Thanks!
Margarita

[Margarita Cabra]

Hi Charlie,
may I ask you how did you deal with the problem? I have the same issue now and don't know what to do.
Thanks!
Margarita

For thos who have this issue, I solved it very easy doing a merge.