MCSN Thursday, 3-Nov-11

From CCE wiki archived
Jump to: navigation, search


Note: I've blocked off next Wednesday, 2:30 to 3:15 during my office hours for anyone in this class who wishes to stop by...


Two main approaches:

  • single network, in which you relate structural and attribute data, as summarized by partitions and vectors, then interpret the results (e.g.: "cohesive subgroups can be related to musical genre, and I think this is why...")
  • multiple networks, in which you compare overall structural attributes and try to understand differences by relation to attributes (e.g. "egonets for punk rock fans are more densely connected than egonets for jazz fans, and I think this is why...")

Main modelling issues:

  • how to model the phenomenon using SNA
    • are we dealing with a one, two, or multiple mode network?
    • is it directed or not? simple or not?
    • how to define "vertex"
      • what does "vertex" mean? are there qualitatively different kinds?
      • how to define attribute information for vertices (partitions and vectors)?
      • arbitrariness management
        • what's the cutoff between vertex and no vertex?
        • where are the network boundaries, and can they be justified?
    • how to define "line"
      • what does "line" mean? are there qualitatively different kinds? arcs or edges?
      • arbitrariness management:
        • what's the cutoff between "line" and "no line"?
        • how to represent strength of relation?
      • how to define attribute information for lines (line values)?

Check some of the ideas I've listed...


  • Organize proposal strictly according to the 5 section form (plus a 6th section for bibliography) so I can see exactly how you're thinking (number them for clarity). This will become the first section of your paper. The other sections will simply expand on these, broadening the context/background, answering the queries, defining the model, applying the method....all you'll have to add is a conclusion section in which you summarize results, critique possible flaws in the model or method, and suggest further directions (queries, methods) for future research. Here again are the 5 sections, plus bibliography:
    • phenomenon, aim, value
    • context/background
    • query
    • SNA model
    • SNA method
    • bibliography
  • Identify phenomenon, assess aim and value. Provide in the broadest of terms, without reference to SNA. You may state a website as the focus of research here - or save that for "Method" below, since presumably the phenomenon (e.g. "musical influence") isn't essentially web-based, but rather is represented on the web. However if you'd rather say that it's the web you're studying, fine too.
  • Contextualize: what does the reader need to know? Be concise but do address such things as: the musical style, people who participate, how they relate to one another. Consider this to be the section in which you clarify all the issues and terms used, so that you can use them freely from now on. Cite general sources on the musical styles and figures, scholarly or even popular sources (e.g. magazine articles) are fine here, especially if you're documenting a contemporary trend. (But please don't use wikipedia, though feel free to use it to find things)
  • Query. Try to formulate the questions first without reference to SNA. These can be a bit more specific than in the opening "aim". It is always motivational and instructive to examine contrasts, make comparisons, lest you get lost in the exploration. For instance, you can examine musical influence in general, but it may be more fruitful to show how influence and gender relate to one another. Comparing male and female: who influences whom? Or contrast an influence analysis across musical styles, showing how each style evinces its own patterns in the influence network.
  • Model. Here's where you bring in SNA for the first time. Show how you intend to interpret the phenomenon in terms of networks so as to create an SNA model. Be sure to define how concepts of "vertex" and "line" (and perhaps vertex value and line value, and related partitions and vectors) can be used to describe your phenomenon. You need to be very precise - how do you represent concepts in network terms?
  • Method. Here you can discuss the data sources (e.g. website), and SNA concepts or Pajek techniques to try out. In your methodology, consider all the network concepts and Pajek techniques we've learned thus far (e.g. density, degree, components, cores, cliques, signed networks, affiliation networks...more is coming). Until you can demonstrate to yourself that a given concept or technique is *not* applicable, leave it in. During the exploratory phase you'll try things out. Don't close off avenues of exploration prematurely. Consider also the visual display of information - what techniques may work best to summarize your results?
  • The bibliography can combine various types of sources:
    • Primary sources - perhaps websites, or newspapers, published datasets such as Statistics Canada, etc.
    • Secondary sources about the music under discussion - whether historical, ethnographic, interpretive, critical, or biographical...
    • General secondary sources (probably from the social sciences) treating concepts of interest, e.g. fame, influence, consumption, marketing, promotion, production - whether in music or more broadly
    • SNA sources. I don't expect you to include many of these but if you can locate some papers that treat a similar subject or issue, include them and do your best to read and assimilate what they contain. Certainly you can all add ESNAP as a prime source, and cite when you deploy a particular technique.
    • Please don't use wikipedia as a source, though feel free to use it to find sources. Use also the INSNA site, and jstor and other library databases in sociology and music.


  • Think about research questions you can answer (you have the needed data) and that lend themselves to SNA techniques.
  • Start creating some networks; playing around with them will help make your research plan more concrete, will suggest further questions, and motivate you to pursue them.
  • Be sure you're really using SNA. Is there a network in your data? If you have only partitions (clustering actors in various ways) or vectors (assigning values to actors), then you're not going to be doing anything other than ordinary statistics. Actors need to be connected to one another beyond simple clustering (perhaps via a second type of vertex - i.e. in a 2-mode network).
  • Importance of comparison: Be sure you can come up with conclusions, at least potential conclusions. This requires that there be potential relationships to interpret, resulting from comparisons. There are basically two ways to go:
  • Comparisons within a single network.
    • Vertices have network properties: which cohesive subgroups they belong to, their degree, how central they are...
    • Vertices have attribute properties, apparently independent of any network - such as "success", "gender", "style, "genre", "place", "instrument", "decade"
    • Each property can be represented either as a partition or as a vector.
    • You can try to 'correlate' (run crosstab tests) these properties.
  • Comparisons across networks, attempting to detect systematic differences in network properties.
    • Develop parallel, independent networks illustrating the same phenomenon (e.g. links from artists to labels), but for different:
      • genres
      • genders
      • instruments
      • decades
      • labels
    • Then compare whole-network properties (centralization, network components, density, average degree).
  • Metrics of artist success (money, fame, quality):
    • sales (quantity, $$$)
    • critical ratings - reviews
    • popular ratings on websites
    • charts
    • measures of fame (youtube hits, newspaper or media references, google hits, google links (try link: keyword), etc.)
  • Identify cohesive subgroups in terms of vertex attributes, such as "musical style" or gender.
  • Refinements in affiliation networks
  • Egonets: the egonet for a particular vertex is the subnet comprising its neighbors and their interconnections.
  • Temporal nets
    • Flows in exclusive affiliation nets (in which an artist, say, is affiliated to one place/label/venue at a time)
    • Diffusion. Observe what happens as a new album is released, say. How does information spread? What patterns are exhibited? Twitter is very useful here as a data source.
    • Changes over time, generally

Your brief presentations

Chapter 5