Training and Events
Multilevel Analysis of Social Network Data
|University of Manchester|
Prof Mark Tranmer
Cathie Marsh Institute
View in Google Maps (M13 9PL)
Claire Spencer, 0161 275 1980, firstname.lastname@example.org
This course is aimed at researchers who are interested in the potential for Multilevel Analysis with Social Network Data. The focus will mainly be about cross-sectional data, and there will be both lecture-based and hands-on components to the course.
The course will explain how multilevel analysis can be useful for the analysis of social network data, given certain research questions relating to it.
The course begins by briefly reviewing multilevel models for the classic “pupils-in-schools” multilevel analysis example, before explaining how the model framework for that example can then be adapted and developed for social network analysis.
The course focus is on two situations for the multilevel analysis of cross-sectional social network data, with theoretical and practical examples in both cases:
1. Multilevel Models for non- or minimal overlapping ego-nets.
An ego-net comprises a focal actor (ego) and the actors to which it is connected (alters). The most obvious example is a person (ego) naming their friends (alters). Another example is a person naming their support network. Such data can be collected through surveys, where friendship or support questions are included, or may exist in administrative data sources. In some situations, the actors may not necessarily be people and could be, say, organisations.
The ego-alter ties in an ego net could be binary e.g the alter to whom ego is tied has a particular characteristic e.g. is a medical professional (=1) or not (=0), or interval ties: e.g how much support time does each alter give ego in a typical week? In a multilevel analysis of non- or minimal overlapping ego net data, these ties are the units of analysis, and the model we use will have an alter and an ego level. Attributes of the ego and alter can be used in this model framework to predict the presence or value of a particular tie between alter and ego. For binary ties a logistic model is used and for interval ties a linear model is used. Moreover, similarity of characteristics of ego and alter can be modelled in the multilevel framework, allowing the association of the similarity (homophily) or difference (heterophily) of ego-alter attributes with the tie values to be assessed, and the strength of associations between attributes and ties can vary from ego-net to ego-net. These features make the multilevel model valuable for a variety of substantive social network studies.
2. Multilevel Models for social network dependencies.
Sometimes in social network analysis the focus is not on modelling social network structure (e.g. with Exponential Random Graph Models (ERGMs)), but on modelling social network dependences. Social network dependencies are the extent to which the responses of individuals are associated with responses of other individuals to whom they are connected. A classic example is the peer-effect model, say for pupils in schools, where the exam score of a pupil is correlated with the exam scores of their friends. In social network analysis, Network Autocorrelation Models (NAMs) are sometimes used to investigate network dependencies in this context. An alternative is to use a multilevel approach via a multiple membership model which allows the extent of network variations in values of the responses to be investigated, possibly in the context of other population groupings, such as local areas, and has several other distinct features from other modelling approaches for networks. Such features make the multilevel approach a valuable in social research for social network dependencies, as will be explained.
Participants should be experienced users of single-level regression models – both linear and logistic. Familiarity of multilevel modelling would be useful but is not essential.
The course will use the software R for the analysis, preparation and modelling of the data. Especially for the multiple membership models, we will also draw on MLwiN, invoked from R via the R2MLwiN package. The practical sessions will include all necessary code for R/R2MLwiN, so that an all-round knowledge of R is not required. Some familiarity with R would however be useful, via for example: Quick-R:www.statmethods.net but is not absolutely essential.
Some people may have used MLwiN before – although MLwiN will not be used directly for the practical work on this course, MLwiN screen shots will be included in the course documentation, for comparison with fitting the models via R/R2MLwiN.
We don’t assume you will read all the material below, but if you did at least look at some of it, it would significantly enhance your learning on the course.
Crossley N, Bellotti E, Edwards G, Everett M, Koskinen J and Tranmer M (2015) Social Network Analysis for Ego-Nets. Sage Publications. [especially Chapter 6].
Multilevel Models for non- or minimal overlapping ego-nets:
Snijders, T., Spreen, M. and Zwaagstra, R. (1995) The Use of Multilevel Modeling for Analysing Personal Networks: Networks of Cocaine Users in an Urban Area, Journal of Quantitative Anthropology 5(2), 85–105.
de Miguel Luken and Tranmer M (2010) Personal Support Networks of Immigrants to Spain: a Multilevel Analysis. Social Networks, 32, no. 4: pages 253-262.
Multilevel Models for social network dependencies:
Tranmer M, Steel D, and Browne W (2014) Multiple Membership Multiple Classification Models for Social Network and Group Dependencies. Journal of the Royal Statistical Society, Series (A), 177, Part 2, pages 1-17.
Multilevel Models overview:
Snijders, T.A.B. and Bosker, R.J. (2012) Multilevel Analysis: An Introduction to Basic and Advanced Multilevel Modeling, 2nd edition, London, Sage Publishers.
R, R2MLwiN, MLwiN web resources:
Quick r website: www.statmethods.net
R studio http://www.rstudio.com
MLwiN (free to UK based academics) http://www.bristol.ac.uk/cmm/software/mlwin/
Intermediate (some prior knowledge)
Website and registration
Analysis of social media, Social media data, Regression Methods, Social Network Analysis, R
Related publications and presentations