types of social network analysis courses

We live in a golden age of network analysis. It’s booming as science and booming as business. This raises questions for the teacher – what course should you teach? A few options:

1. Bare bones: A course designed for folks with little to no mathematical background. You would teach descriptive stats, visualization, and applications.
2. Stats+/Models : In this course, you’d assume some basic background. Maybe micro for econ students or stats for other social science students. Then, you’d dig deep into different centrality measures, power laws, clustering/community detection, etc. A follow up course would deal with p*, ERGM, Sieana and other advanced issues.
3. Programmers: Here, you’d lightly gloss over the math and proofs and instead focus on how to scrape the net for data, how to write simulations, and how to manipulate big data sets.
4. Elite stats: This is for a very small number of students in math, stats, or econometrics. It would be exclusively proofs of fairly advanced issues (like the graph models underlying p*).

Currently, I teach a course for sociology seniors between 1 and 2. I get soc students, a handful from econ/psych/poli sci, and one or two informatics students. I also get one or two grad students. The elite soc programs, where students often have science backgrounds or simply a lot of mojo, are now seeing Programmers courses. Old school networks courses (a la Ed Laumann or John Padgett at Chicago) offer a version of #2. Elite stats is exceptionally rare in that if students are that advanced, they can often read the papers themselves. Add your own comments about networks education.

50+ chapters of grad skool advice goodness: Grad Skool Rulz ($2!!!!)/From Black Power/Party in the Street

more tweets, more votes: social media and causation

This week, the group Political Bots wrote the following tweet and cited More Tweets, More Votes in support:

why do politicians buy bots? "statistically significant association" more tweets more votes U.S. House by @fabiorojas http://t.co/gS6FjEN9Ig

— politicalbots.org (@polbots) September 2, 2015

The claim, I believe, is that politicians purchase bots (automated spamming Twitter accounts) because they believe that more presence on media leads to a higher vote tally.

In presenting these results, we were very careful to avoid saying that there is a causal relationship between social media mentions and voting:

These results indicate that the “buzz” or public discussion about a candidate on social media can be used as an indicator of voter behavior.

And:

Known as the Pollyana hypothesis, this finding implies that the relative over-representation of a word within a corpus of text may indicate that it signifies something that is viewed in a relatively positive manner. Another possible explanation might be that strong candidates attract more attention from both supporters and opponents. Specifically, individuals may be more likely to attack or discuss disliked candidates who are perceived as being strong or as having a high likelihood of winning.

In other words, we went to great efforts to suggest that social media is a “thermometer,” not a cause of election outcomes.

Now, it might be fascinating to find that politicians are changing behavior in response to our paper. It *might* be the case that when politicians believe in a causal effect, they increase spending on social media. Even then, it doesn’t show a causal effect of social media. It is actually more evidence for the “thermometer” theory. Politicians who have money to spend on social media campaigns are strong candidates and strong candidates tend to get more votes. I appreciate the discussion of social media and election outcomes, but so far, I think the evidence is that there is not a casual effect.

50+ chapters of grad skool advice goodness: Grad Skool Rulz ($2!!!!)/From Black Power/Party in the Street

relational styles in micro-finance

A long standing issue in network analysis is the analysis of when people initiate and maintain relationships. Rodrigo Canales and Jason Greenberg have a forthcoming Management Science paper that uses data from interaction between micro-finance loan officers and clients to establish that interactional style is one of the big drivers of relationships. From the abstract:

Social scientists have long considered what mechanisms underlie repeated exchange. Three mechanisms have garnered the majority of this attention: Formal contracts, relational contracts, and relationally embedded social ties. Although each mechanism has its virtues, all three exhibit a common limitation: An inability to fully explain the continuation and stability of inter-temporal exchange between individuals and organizations in the face of change. Drawing on extensive quantitative data on approximately 450,000 microfinance loans made by an MFI in Mexico from 2004-2008 that include random assignment of loan officers, this research proposes the concept of “relational styles” to help explain how repeated exchange is possible in the face of personnel change. We define relational styles as systematically reoccurring patterns of interaction employed by social actors within and across exchange relationships — in this paper, between microfinance clients and loan officers. We show that relational styles that are consistent facilitate a clear understanding of expectations and thus exchange. We also demonstrate that consistency in the relational styles followed by successive loan officers mitigates the negative impact of a broken loan officer-client tie. This paper thus proposes and empirically tests a social mechanism based on relational styles that often accompanies relational embeddedness, but may also serve as a partial substitute for it.

Check it out!!!

50+ chapters of grad skool advice goodness: Grad Skool Rulz ($2!!!!)/From Black Power/Party in the Street

science as a giant conscious hive mind

In a new article in Social Networks, Feng Shi, Jacob Foster, and James Evans argued that the complexity and diversity of scientific semantic networks creates very high rates of innovation. From Weaving the fabric of science: Dynamic network models of science’s unfolding structure:

Science is a complex system. Building on Latour’s actor network theory, we model published science as a dynamic hypergraph and explore how this fabric provides a substrate for future scientific discovery. Using millions of abstracts from MEDLINE, we show that the network distance between biomedical things (i.e., people, methods, diseases, chemicals) is surprisingly small. We then show how science moves from questions answered in one year to problems investigated in the next through a weighted random walk model. Our analysis reveals intriguing modal dispositions in the way biomedical science evolves: methods play a bridging role and things of one type connect through things of another. This has the methodological implication that adding more node types to network models of science and other creative domains will likely lead to a superlinear increase in prediction and understanding.

Bringing soc of science and network analysis together. Love it.

50+ chapters of grad skool advice goodness: Grad Skool Rulz ($2!!!!)/From Black Power/Party in the Street