Data sources and types

Hanneman and Riddle, chapters 1, 2, and 6.

• Introduction: What's different about social network data?
  • Conventional data as actor by trait: rectangular
  • Network data as actor by actor, one layer for each relation: square
  • Rectangular network data: actors by events
  • Bi-Partite network data: multiple types of actors
• Nodes
  • Populations, samples, and boundaries
  • networkers usually study populations
  • sometimes samples of populations are used
  • drawing population boundaries and getting closure can be a problem
• Modality and levels of analysis
  • single mode versus multi-mode networks
  • most studies analyze one or two levels of embeddedness
• Relations
  • actors usually have many types of ties
  • material and informational flows
  • bonded ties or co-occurrence
• Sampling ties
  • Full network methods
  • Snowball methods
  • Ego centric networks (with alter connections)
  • Ego only "networks"
• Multiple relations
  • collapsing and indexing data on multiple relations
  • role algebras and other approaches to multiplexity
• Scales of measurement
  • nominal or binary data (most common)
  • multiple category nominal data
  • ordinal data
  • special case: the signed graph and balance
  • ordinal scales of strength of ties
  • interval (valued) data
• A note on statistics and social network data
  • networkers often don't use samples
  • when they do use samples, the observations aren't independent
  • statistical methods are sometimes used to analyze network simulations

1. Explain the difference between "rectangular" data of actors by traits versus "square" data of actors by actors.

2. What is a "co-occurance matrix?"

3. What is a multi-modal network? Why do network analysts sometimes include more than one mode in the data sets they collect?

4. Social actors may share have many kinds of relations, and share many statuses or bonded relations. How do we decide what ties to measure?

5. Explain the differences between full-network, snowball, ego with alters, and ego only network sampling.

6. Explain the differences between binary, multi-category nominal, grouped ordinal, full ordinal, and interval levels of measurement of tie strength.

7. Network analysts don't use statistics very frequently. Why is this?

Application Questions

1. Think of the articles we read in the first part of the course. For at least two articles, identify the population boundaries of actors, the relation or relation among the actors that was measured, what level of measurement was used, and what kind of sampling of ties was done.

2. Suppose we had a population of 10 persons, 5 male and 5 female. Show how a "conventional" actor by trait data analysis would array data such as these. Now, suppose that you are a social network analyst, and you believe that sharing the trait of having the same sex constitutes a tie between actors. Show how you would array the data on the ten actors and their sexes for a network analysis. Note that many (but probably not all) data analysis problems can be represented in more than one way -- depending on what you want to analyze or emphasize.

3. I am interested in studying why some elementary school students do better in math than others. I think that these differences may have something to do with individual differences, but may also be due to differences between class rooms in a school, between schools in a district, and between school districts. If I collected data on information flows among children, classrooms, schools, and districts, I would have a multi-modal network. How might you represent such an array of data?

4. Suppose I am studying support for a political candidate in a neighborhood. I think that ties between neighbors might affect political attitudes and who supports the candidate. What relations among the neighbors might I want to collect data about?

5. Think of some problem of interest to yourself that involves measuring a single kind of tie between actors in a population. Describe how you would measure the strength of the tie of interest at the nominal, ordinal, and interval level. What would your preference be if you were to actually do the study, and why?