The hidden structure of (academic) organizations

All organizations have a ‘deep’ hidden structure based on the social interactions among its members which might or might not coincide with the official formal one. University departments are no exception – if anything, the informal alliances, affinities, and allegiances within academic departments are only too visible and salient.

Network analysis provides one way of visualizing and exploring the ‘deep’ organizational structure. In order to learn how to visualize small networks with R, I collected data on the social interactions within my own department and plugged the dataset in R (igraph package) to get the plot below. The figure shows the social network of my institute based on the co-supervision of student dissertations (each Master thesis has a supervisor who selects a so-called ‘second’ reader who reviews the draft and the two supervisors examine the student during the defence). So each link between nodes (people) is based on one joint supervision of a student. The total number of links (edges) is 264 which covers (approximately) all dissertations defended over the last year. In this version of the graph, the people are represented only by numbers but in the full version the actual names of people are plotted, the links are directional, and additional info (like the grade of the thesis) can be incorporated.

Altogether, the organization appears surprisingly well-integrated. Most ‘outsiders’ and most weakly-connected ‘islands’ are either occasional external readers, or new colleagues being ‘socialized’ into the organization. Obviously, some people are more ‘central’ in the sense of connecting to a more diverse set of people, while others serve as boundary-spanners reaching to people who would otherwise remain unconnected to the core.  I find the figure intellectually and aesthetically pleasing (given that it is generated with two lines of code) and perhaps a more thorough analysis of the network can be useful in organizational management as well.

Facebook does randomized experiments to study social interactions

Facebook has a Data Science Team. And here is what they do:

Eytan Bakshy […] wanted to learn whether our actions on Facebook are mainly influenced by those of our close friends, who are likely to have similar tastes. […] So he messed with how Facebook operated for a quarter of a billion users. Over a seven-week period, the 76 million links that those users shared with each other were logged. Then, on 219 million randomly chosen occasions, Facebook prevented someone from seeing a link shared by a friend. Hiding links this way created a control group so that Bakshy could assess how often people end up promoting the same links because they have similar information sources and interests  [link to source at Technology Review].

It must be great (and a great challenge) to have access to all the data Facebook and use it to answer questions that are relevant not only for the immediate business objectives of the company. In the words of the Data Science Team leader:

“The biggest challenges Facebook has to solve are the same challenges that social science has.” Those challenges include understanding why some ideas or fashions spread from a few individuals to become universal and others don’t, or to what extent a person’s future actions are a product of past communication with friends.

Cool! These statements might make for a good discussion about the ethics of doing social science research inside and outside academica as well.

Predicting the votes of judges

Here is a (short) and interesting paper that uses an innovative approach to predict the votes of the US Supreme Court:

Successful attempts to predict judges’ votes shed light into how legal decisions are made and, ultimately, into the behavior and evolution of the judiciary. Here, we investigate to what extent it is possible to make predictions of a justice’s vote based on the other justices’ votes in the same case. For our predictions, we use models and methods that have been developed to uncover hidden associations between actors in complex social networks. We show that these methods are more accurate at predicting justice’s votes than forecasts made by legal experts and by algorithms that take into consideration the content of the cases. We argue that, within our framework, high predictability is a quantitative proxy for stable justice (and case) blocks, which probably reflect stable a priori attitudes toward the law. We find that U.S. Supreme Court justice votes are more predictable than one would expect from an ideal court composed of perfectly independent justices. Deviations from ideal behavior are most apparent in divided 5–4 decisions, where justice blocks seem to be most stable. Moreover, we find evidence that justice predictability decreased during the 50-year period spanning from the Warren Court to the Rehnquist Court, and that aggregate court predictability has been significantly lower during Democratic presidencies. More broadly, our results show that it is possible to use methods developed for the analysis of complex social networks to quantitatively investigate historical questions related to political decision-making.

While I have my reservations whether “trying to predict the behavior of judges, one can get insights into how legal decisions are truly made”, exercises in predicting outcomes are interesting in their own right. And this paper appears to hit the target: its predictive success rate is 83% vs. the less-than-70% success rate of existing approaches based on expert opinions and statistical models of case characteristics. Note however that each individual vote is predicted with information about how the other judges have voted on that same case which, if the votes are announced simultaneously, doesn’t provide you with any leverage in actually predicting the outcome of a case.

P.S. What is this penchant that the real scientific journals (e.g.PLoS) have for social science research based on agent-based modeling or network theory?

Concentration of control in the global economy

All conspiracy theorists know that the global economy is concentrated in the hands of a few. But even they will be blown away by this paper which maps the network of global corporate ownership and control. Here is the (somewhat understated) abstract:

“The structure of the control network of transnational corporations affects global market competition and financial stability. So far, only small national samples were studied and there was no appropriate methodology to assess control globally. We present the first investigation of the architecture of the international ownership network, along with the computation of the control held by each global player. We find that transnational corporations form a giant bow-tie structure and that a large portion of control flows to a small tightly-knit core of financial institutions. This core can be seen as an economic “super-entity” that raises new important issues both for researchers and policy makers.”  (Vitali, Glattfelder and Battiston)

Some of the findings:
– almost 40% of the economic value of transnational companies in the world is in the hands of a group of 147 tightly-interconnected companies “which has almost full control over itself” (p.6)
– “[N]etwork control is much more unequally distributed than wealth…[T]he top ranked actors hold a control ten times bigger than what could be expected based on their wealth” (p.6)
10 companies control 20% of the network; 50 companies control 40% of the network (!)
– 35 of these 50 companies belong to a strongly connected core, meaning that they are all “tied together in an extremely entangled web of control” through co-ownerships (p.32)
77% (463 006) of the firms in the entire network belong to a single connected component [formally, in a connected component all firms can reach each other along the paths of the network]. The second largest connected component has only 230 firms.  

Here is the map of the core of the core of the network itself; not very informative as such but beautiful nonetheless:  (Superconnected companies are red, very connected companies are yellow)


(Image: PLoS One, via New Scientist)
 
This is the first paper to include indirect and weighted control paths in constructing the global economy network and it introduces a new method for measuring control that is suitable for such complex networks. Although quite technical, the paper does a remarkable job of walking the reader step-by-step through the analysis. New Scientist has a less-technical presentation of the research here.

The implications of this work for the stability of the economy and competition should be quite obvious, but the authors (all from ETH Zurich) also explicitly discuss them in the paper. One can only hope that economic policy makers and politicians take note.