I should probably start this with a disclaimer—this is not intended to be an in-depth description of the use of social networks for research. It is simply my rehashing of a departmental seminar that has been filtered through my statistics-challenged brain (I finished all my college math in high school and did my undergrad studies in music so it has literally been years since I studied math with any intensity)
First of all, why do we even want to study social networks? Networks can give different views of a subject than standard data alone can. Networks allow us to see how structure matters, separate metaphors and concrete differences, attributes vs. relations, study how an individual's choices constrained by the social structure they reside in, and compare micro and macro views of the same data.
Visualizing Network Data
Network data can be viewed as both a graph and matrix. Graphs are made by plotting nodes and denoting connections using lines. In one type of graph, known as spring and bedding, nodes are pulled closer by the presence of ties. A matrix corresponds to a graph and numerically describes connections between nodes. (See my extremely simplified, rather rough examples of a graph and matrix below)
N1 | N2 | N3 | Etc… | |
N1 | - | 1 | 0 | |
N2 | 0 | - | 1 | |
N3 | 1 | 1 | - | |
Etc… |
1=connection between nodes
0=no connection
After creating visualizations of the data we can then describe and analyze it. The different ways of describing are by composition and structure. Composition would focus on the alters (or nodes) that make up the bedding of the data. Structure can be looked at different ways. By density (less dense areas vs. more dense areas), modal degree, distance between nodes, similarities or patterns of ties.
Analyzing Network Data
There are three basic types of data analysis
Dyadic analysis involves gathering information from a pair of alters
Egocentric analysis is the collection of data around a single node
Complete analysis looks at all the information from everyone
The tools that can be used to assist in these types of analysis are visualizations, cultures and subgroups, the network as a dependent variable (this looks at the probability of tie formation in the network), network as the independent variable (this looks as contagion and influence including the flow of information, power/authority, cohesion/solidarity, and competition/comparison)
Networks, like any research, are not perfect. One common issue is the homophily or selection problem. The question with this problem is "did the friendship form on the basis of the variable being studied?" There is a tendency for people with same characteristics to become friends so this is a very real problem. Another issue is confounds—are the results related to a larger contet (e.g. environment) that is shared by both alters?
Health Implications
This is where network research appeals to public health practitioners. And I thought this portion of the lecture helped me understand the previous intro to networks. Following is a brief review of actual research using each of the three basic types of network analysis.
Dr. Paik described his research utilizing a longitudinal study of adolescent health (ADO Health)—Wave 1 occurred in 1994-95, with subsequent waves in 96, 01, 04
Examples of the different types of data collection used:
Dyadic data—interpersonal violence
Egocentric—sexual concurrency, chlamydia infections
Complete—peer effects of nonromantic sex
Dyadic data collection to evaluate interpersonal violence:
The independent variable was the partner's prior violence from Wave I. The dependent variable was the victimization of the alter in Wave II. A positive correlation was seen, which essentially means that a partner with a history of violence is more likely to continue to be violent.
Egocentric data collection to determine the connection between sexual concurrency and chlamydia infections:
The dependent variable here is a chlamydia infection. The independent variable was the presence of concurrent (multiple) sexual partners. A positive correlation was seen between these variables. As many people would assume, more prior partners increased infections, but having concurrent sexual partners had an even larger increase in infections.
Using the network to find patterns regarding nonromantic sex and delinquency:
The dependent variable in this example is having nonromantic sex during Wave I (or "hooking up" in laymen's terms). Two different analyses occurred using the mean of friends' delinquency as the first independent variable and the network centrality of the respondent weighted by the centrality of his/her friends as the second independent variable.
As one might assume: the respondents' delinquency had a positive effect on the occurrence of hook-ups as did the mean of the friends' delinquency. Simply looking at the friends' centrality showed no effect, but weighting centrality with delinquency showed two differences: having delinquent and central friends increased the odds that a respondent would hook-up but having delinquent, non-central friends actually decreased those odds.
Further analysis?
Are hook-ups contagious as the data seems to suggest?
To look at this they looked at the ties found between those who have never hooked up, ties found between those who had hooked up and those who had not hooked up, and the ties found between people who had had hookups.
Hook-ups did appear to be "contagious". These findings can then be put to use from a public health perspective.
Admittedly, this overview of social networks is extremely simplified and possibly just plain wrong, but if it has piqued your interest in networks Dr. Paik suggests a few books that are helpful for learning more.
Networks and Health by Tom Valente
Networks an introduction by Mark Newman
Networks, Crowds, Markets by Easley/Kleinberg
Coming up—a brief interview with Dr. Paik
~L
Posts
No comments:
Post a Comment