Working With Two Mode Social Network Data in R

This video demonstrates two methods for importing and transforming 2-mode network data (also known as bipartite networks or affiliation matrices) in the open-source research program R. Method #1 involves direct input of an affiliation matrix into R, a method good for relatively small matrices. For larger matrices, it's easier to enter affiliation data into a spreadsheet and import a .csv (comma-delimited) file into R. One hypothetical affiliation matrix and one actual affiliation matrix of corporate board interlocks are used to illustrate the development of R scripts for 2-mode network research. This video was created for the University of Maine at Augusta undergraduate social science program.

[@felixoffei1867]

Thanks, Professor Cook. It has been very helpful in the project research work.

[@eslamamin9129]

Excellent!!

[@minyanshao8178]

Really useful！ THANKS!

[@thespokesman3017]

Hello professor, great video! I was wondering: is it possible to make a one-mode social network, similar to the one you show at 29:41, where you show how individuals are connected through affiliations, but color code them based on gender. in short: is it possible to add a third attribute to the social network?

[@JamesCookUMA]

The short answer is yes! The key is to add a vertex that contains the attributes of nodes (in this case, the gender of nodes). See igraph.org/r/doc/vertex_attr-set.html for information about how to do that.

[@hardbeeda90]

Thanks James for your helpful videos about social network analysis. I have a silly question for which I hope your answer may enlighten me. I am very new to R but I have seen suggestion to convert the 2-mode network into a 1-mode network to calculate centrality measures at the NODE/Individual level. If I want to calculate centrality measures at the MODE level, e.g: board/corporate level just like in your example, what should I do in this case? Your answer would be greatly appreciated!

[@JamesCookUMA]

Hello! It's not a silly question at all. Once you've converted a 2-mode network into two 1-mode networks (one an individual-x-individual network and the other a corporate board-x-corporate board network), you can calculate centrality measures at the NODE level, whether, the node is an individual or the node is a corporate board. In principle, it's that simple. But in practice, notice that in the conversion of a 2-mode network into two 1-mode networks you have valued data in the cells (not just a 1 for "tie present" or 0 for "tie absent" but a number for strength of tie). One solution to this is to dichotomize your data (set all numbers above a certain threshold to 1, set all numbers at or below that threshold to 0), and then to apply a centrality measure. Another solution is to adopt a different centrality measure that accommodates valued data. See Freeman, Borgatti and White 1991 for a classic example of the latter approach: moreno.ss.uci.edu/54.pdf .

[@24shivendra]

Hello Professor ! I am a student from Italy. Your videos have been immensely useful for my thesis. I am analyzing the network of directors of the largest 800 corporations of Europe. I have created an affiliation matrix of dimension 13094 x 800. Using R, I have obtain a visualization but because of the huge amount of data, it appears very cluttered and is visually very difficult to make sense. What would you suggest ?

[@JamesCookUMA]

I hear you! Network graphs can be very cluttered! Have you turned your affiliation matrix into an 800x800 adjacency matrix (director overlaps?). Assuming so, I'd suggest either: A) not relying on visualization with that many nodes and carrying out a quantitative analysis instead, or B) carrying out a combination of selecting the most important nodes in the graph by some metric such as centrality and visually representing certain categorical characteristics of the corporations. But that's just offhand without knowing too much about the substantive aspects of your interest.

[@24shivendra]

Thank you very much. Taking your input into consideration, coupled with the computation of betweenness, degree and closeness centrality of all the directors and corporations, I can draw a conclusion about the important actors and groups in this network. I appreciate your valuable contribution. Although, I tried to use the ZOOM package in R, to zoom into my 1-mode director network, but it didnt seem to work.

[@savinglives3482]

hello professor, the video is really helpful. please can you send me your email

[@JamesCookUMA]

My e-mail is [e-mail]