Research Methods in Social Network Analysis
Research Methods in Social Network Analysis
Research
Methodology
Linton C. Freeman, Douglas R. White and A. Kimball Romney are professors in the Pro-
gram in Social Network Analysis at the University of California, Irvine.
www.routledge.com
Safety Area: All Text, Logos & Barcode should remain inside the Pink Dotted Lines
Bleed Area: All Backgrounds should extend to, but not past, the Blue Dotted Lines
Research
Methods in
Social Network
Analysis
Q Taylor & Francis
~ Taylor & Francis Group
http://taylora n dfra ncis.com
Research
Methods in
Social Network
Analysis
Edited by
Linton C. Freeman
Douglas R. White
A. Kimball Romney
I~ ~~O~~~;~~;UP
LONDON AND NEW YORK
Originally published in 1989 by George Mason University Press.
Notice:
Product or corporate names may be trademarks or registered
trademarks, and are used only for identification and explanation
without intent to infringe.
[HM131.R465 1991]
300-dc20 91-23740
CIP
ISBN 13: 978-1-56000-569-8 (pbk)
T ABLE OF CONTENTS
Introduction ....................................................................................... .1
I. Network Representations
Chapter 1. Social Networks and the Structure Experiment
Linton C. Freeman ...........................................................11
David Strauss
Department of Statistics
University of California, Riverside
Douglas R. White
School of Social Sciences
University of California, Irvine
Q Taylor & Francis
~ Taylor & Francis Group
http://taylora n dfra ncis.com
INTRODUCTION
I. Network Representations
equivalence to social roles. The chapters in this section deal with all four
concepts of position.
Before discussing the chapters in this section, we might note that
Foster and Seidman, in Chapter 2, provide a generalization of the ordinary
concept of cliques and maximally complete subgraphs. They also provide a
general discussion of the concept of position in a network, in graph-theoretic
terms. A subgraph has an intrinsic structural feature as a function of its self-
contained relations. It has a positional characteristic as a function both of
internal relations and those with outside elements. A clique or cluster is thus
defined positionally. Structural equivalence and regular equivalence, as role
concepts, are also defmed positionally.
In Chapter 10, Mitchell applies both cliquing and structural
equivalence algorithms to Kapferer's Zambian data, before and after a labor
dispute, on shop-floor networks. He examines the extent to which these
analyses, as well as graph theoretic measures based in reachability and
distance, support and refine Kapferer's hypotheses about the structure and
shifts of patterns in the network.
In Chapter 11 a new algorithm (MAPCLUS) for finding overlapping
subsets is applied to matrices of structural equivalence coefficients from two
classic sociometric data sets: the Bank Wiring Group, and the Sampson
Monastery. Arabie and Carroll's results are similar in spirit to the
representations discussed in Chapter 2 for overlapping subsets and those of
Chapter 10 on "islands." The algorithm could be applied to any of the three
main types of positional problems above: (1) to find overlapping cliques,
using as input the raw relations of a symmetric graph; (2) to fmd overlapping
positions of structural equivalence, as exemplified here; or (3) to find
overlapping roles, using as input the regular equivalence coefficients of
Chapter 13.
Arabie and Carroll also note the difficulties of the classic clique-
finding problem of finding maximally complete subgroups of a graph. The
problem is NP-complete, and is thus likely to continue generating a plethora
of competing algorithms. MAPCLUS, on the other hand, uses a
computationally tractable procedure. It differs from other clustering
techniques in allowing implicit overlap, and providing explicit tests of
goodness-of-fit.
Chapter 12 examines the problem of finding non-overlapping clusters
of structurally equivalent points in a network. In the resulting blockmodel of
patterns of relatedness between sets of equivalent points, Heil shows how to
measure goodness-of-fit as a departure-towards high or low density
8 Introduction
blocs-from expected density under the null hypothesis. Using the fit
measure, he develops a procedure (COBLOC) for evaluating successive
agglomerations of points, resulting in a family of blockmodels. He
exemplifies the approach using the Bank Wiring Room data and shows
improvements in the COBLOC approach as compared to earlier
blockmodeling methods.
Chapter 13 takes the third approach to positions in a network, based on
"global" role structures. Reitz and White generalize the blockmodeling
approach to capture broader intuitions about the nature of such roles. They
develop two new types of blockmodels based on positional equivalences that
capture patterns of single-link and multiplex connectivities so that positions
in the network correspond to actor/counterpart relations in a system of
interconnected roles. Unlike the structural equivalence approach, they do
not require that equivalent actors are identically related to other actors, but
only that they have the same relations with some actors in counterpart
equivalence sets.
Thus, Reitz and White extend the network representation developed in
Chapter 1, but introduce a number of new structural concepts and defmitions
to represent the roles occupied by actors. They follow the strategy of
Chapter 2, moreover, in drawing on anthropological research on role
structures and attempting to formalize this richness in mathematical terms.
They present a number of new theorems about the hierarchical ordering of
positional equivalences (of types 2, 3, and 4 above), their uniqueness
properties, and connectivity properties. They show the implications of each
type of equivalence for inducing homomorphic images of the semigroup of
relations on a graph. This part of the chapter is closely related to Chapters 7
and 8, dealing with relational semigroups. In connection with Bonacich's
question, Reitz and White completely specify the conditions for graph homo -
morphisms to induce semigroup homomorphisms or isomorphisms, and they
provide tools for opening up new lines of research on graph and semigroup
structure.
Reitz and White introduce measures of regular equivalence and an
algorithm for finding regular blockmodels based on regular equivalence
coefficients. Their example shows the difference between the abstract
approach to role regularity and the approach based on structural equivalence.
Chapter 14 provides a probabilistic model for evaluating both local
and global aspects of role structure. Marsden asks three types of questions
about roles or subgroups: (1) what are the internal relations within the
category; (2) what are the types of relations by which category members are
connected to other members of the total network (e.g., isolate, receiver,
Social Network Analysis 9