Bodong Chen

Crisscross Landscapes

Notes: Agneessens, F., & Everett, M. G. (2013). Introduction to the special issue on advances in two-mode social networks



Citekey: @Agneessens2013-wp

Agneessens, F., & Everett, M. G. (2013). Introduction to the special issue on advances in two-mode social networks. Social Networks, 35(2), 145–147.






Two-mode data was collected by Davis et al. (1941) in theirclassic study on theparticipation of Southern women in social events and later by a variety of researchers studying interlocking directorates from the 1970s onwards (e.g., Stokman et al., 1985). (p. 1)

Breiger (1974) gave a systematic account of how to analyze such networks and introduced the idea of duality between persons and groups. (p. 1)

Breiger’s approach, which we now often refer to as the projection or con- version approach was the favored technique, but it is now more common to analyzethe datadirectly as a bi-partite graphand make adjustments to account for the fact that some interactions are not possible. (p. 1)

These adjustments were first systematically examined by Borgatti and Everett (1997). There has been an increasing interest in two-mode networks in recent years mainly because there are many instances where the data is of this form. (p. 1)

The next paper is by Everett and Borgatti (2013) and this paper challenges one of the basic statements often quoted by the net- work community that projecting two-mode data means that there isa loss ofdata.Theyacknowledge thatif onlyone projection iscon- sidered, or if the resultant projections aredichotomized then thisis true. But they contend that ifboth projections areused in any anal- ysis and these are not dichotomized that – except in a few very rare cases – there is no loss of data. (p. 1)

This special issue arose partly from a meeting specializing on two-mode data held at the VU University Amsterdam in October 2009, organized by Filip Agneessens, Peter Groenewegen and Ger- hard van de Bunt. (p. 1)

A series of subsequent papers consider statistical models for two-mode networks that try to explain the overall structure of a network by considering local processes. Both exponential random graph models (ERGMs, also referred to as p* models) (Wasseman and Pattison, 1996; Lusher et al., 2013) and stochastic actor- based models (SIENA models) (Snijders, 2001; Snijders et al., 2010) have classically focused on explaining the structure of one-mode networks by considering local forces, such as reciprocity, transitiv- ity and homophily. (p. 1)

Opsahl (2013) examines the issue of defin- ing clustering coefficients for two-mode data (p. 1)

Although ERGMs for two-mode network data were first pro- posed by Skvoretz and Faust (1999), they were only recently extended to include attributes and clustering (e.g., Agneessens and Roose, 2008; Wang et al., 2009). (p. 1)

The third paper by Kleinnijenhuis and de Nooy (2013) has at its heart a more complicated network structure. (p. 1)

complicated bythe fact that the edges are valued with a measure of how extreme a party is (p. 1)

A second application of two-mode ERGMs by Conaldi and Lomi (2013) considers the local self-organizing principle of free/open source software (F/OSS) projects. By considering the bug-fixing of voluntary contributors as a two-mode network, they show how, in the absence of a formal structure, coordination can come about. (p. 2)

Another issue not dealt with so far concerns multiplex two- modenetworks.Inmany cases differentrelationsbetweendifferent modes might exist. (p. 2)

another important issue relates to both the design of the data collection and the boundary specification problem in two-mode networks. Defining who is part of the network and who is out- side the network is often difficult. However, with two-mode data the additional problem arises that generally the boundary for one of the modes is defined by the boundary specified for the other mode. (p. 2)

In another study, Zhu et al. (2013) consider what motivates people in online games to join a team. They study the emer- gence of teams in a Massively Multiplayer Online Role-Playing Game (MMORPG) to understand the self-assembling mechanism of project team. Their results support the idea that homophily mechanisms play an important role in the choice of other team members. However, they also show that complementary expertise is an important reason for joining a team, as is the similarity in status (skill level) of the team members. (p. 2)

Snijders et al. (2013) propose a SIENA framework for studying the evolu- tionof thetwo-modenetwork andthe co-evolution withone-mode networks. Studying the employment preferences of MBA students over time as a two-mode network, their study not only illus- trates the importance of looking at the evolution of this two-mode network (cf. Koskinen and Edling, 2012), but it also shows the importance of considering the co-evolution with the one-mode friendship and advice relations among these MBA students. (p. 2)

Hence, in many cases the delineation of the border of a two-mode network isoften more arbitrary and requires considerable attention. (p. 2)

there are a number of areas in which very lit- tle work has been done. First of all, all the examples in this issue were undirected. (p. 2)

None of the papers really engage with multi-mode networks in which there are more thantwo modes. Networks such as criminals bycrime byvictim are an example and yet few such datasets are collected and probably as a consequence even fewer specialist methods exist. (p. 2)