Thesis: 5.5 Conclusion and discussion
tnet represents a step forward towards incorporating two aspects into mainstream network analysis, namely weight of ties and network evolution. First, the software package enables researchers to conduct a structural analysis of weighted networks. This analysis is not limited to a description of weighted networks (e.g., Panzarasa et al., 2009), but can also be used for investigating the impact of network structure on performance (Ahuja, 2000; Panzarasa and Opsahl, 2007). The importance of using the richness encoded in the weight of ties was highlighted by Newman (2004b) who showed that different authors had the highest closeness score when Freeman’s (1978) binary measure and a generalised version (Newman, 2001c) were applied to a coauthorship network.
Second, tnet provides a simple and methodologically sound framework for studying different growth mechanisms in both binary and weighted longitudinal networks. This type of analysis can produce new findings. For example, it is surprising to find that the number of common friends does not significantly increase the probability of forming a tie in the online social network used in Chapter 4. By making the function available, this has the potential of enabling practitioners to make better choices and formulate appropriate strategies and policies. In particular, it might help to calibrate algorithms like the ones used by Facebook’s service People You May Know.
The software package is not without limitations. First, there are many more network measures that have already been generalised to weighted networks, but not yet included in the package, e.g. the betweenness measure by Freeman et al. (1991). By incorporating these measures, the relevancy of tnet would undoubtedly increase. Second, even though functions have been checked, we cannot rule out bugs. Third, this package is meant to become a truly open-source project with many contributors. However, we have not yet had the opportunity to invite others to join us as it has just become public.
The already existing functions are not without limitations either. For example, networks analysed by the
tnet.growth.clogit-function cannot contain negative ties, i.e. the weakening or severing of ties. Currently, we are working on including this feature. The incorporation of the
add_window_to_longitudinal_data-function forms an integral part of this effort, which allows for the inclusion of a smoothing window to a longitudinal dataset. Moreover, we are also adding the option of including specific node-level variable. For example, if the gender of people in a social network is known, a dummy variable signalling whether the creator node is female or male could be included in the regression.