Social network analysis

These analyses focused on understanding the distribution of power, existence of coalitions and the presence of key social structures in an information exchange network. Studies that use SNA usually focus on relationships at the individual level. I acknowledge that studying individual rather than organisational relationships provides a more nuanced picture of networks, but such networks are usually more dynamic as well. One of the criticisms of SNA is precisely that its scope is significantly limited in time (Bodin and Prell 2011). Considering that the last MPA declarations in Tasmania were in 2009 (see Chapter 4), and that many individuals who participated were no longer involved, I chose to focus my analysis at the organisational level, rather than the individual, as organisations tend to remain stable for longer. Considering the scope of the study described in Chapter 1 and in Section 3.3.1 of this chapter, key organisations were those with a role at the Tasmanian State level.

The method of eliciting relationships has important consequences on the resulting networks. To discover network relationships, I chose what Doreian and Woodard (1992) termed “expanding sampling”. In this method, participants are presented with a list of actors, but they can add organisations to this list if they consider them relevant. In this study, participants were presented with a list of key actors to collect network data (Appendix 5 and questions 4 to 7 in appendix 4). This list was generated following the procedure explained in section 2.3.2. Question 4 inquired about the relationships with other organisations to exchange information; question 5 asked participants to rank actors according to their influence level, while question 6 asked which sources of power were used by highly influential actors. Participants were also asked if they were aware of the existence of any coalitions regarding marine conservation issues (question 7). If more than three people identified additional organisations from those in the original list, these were also invited to participate in an interview (to answer both qualitative questions from section 3.3.5, and SNA questions). Two such organisations were included in the sample, and two answered to the request saying that they did not have an interest in Tasmanian MPAs. For cognitive methods (information provided by a third party, rather than reporting about their own relationships) all actors in the list were included. Cognitive analyses included Hubs and Authorities and Social Cognitive Mapping (see below). For the information exchange network and exponential random graph models (ERGMs - see below), only interviewed organisations were included. Multi-sector groups, such as advisory committees, were excluded from the analysis. I made this decision because no representative would have been able to provide a unified depiction of the group’s networks. In a similar way, roles, such as the Minister for

Primary Industries, fluctuate too much to provide a reliable idea of social networks, unless these are very specific in time.

Most SNA data were analysed with UCINET 6 (Borgatti et al. 2002) and the PNet software was used for estimation of ERGMs (Wang et al. 2009). Results on the distribution of influence were analysed in UCINET 6 to obtain the Hubs and Authorities score for each stakeholder, E-I indexes (explained below), to calculate centrality measures (Freeman 1979) and to visualise the information exchange network.

As mentioned in Chapter 2, power needs to be organized in a network to have an impact, and SNA therefore provides a useful method to study power (Domhoff 2014). In this thesis I used the Hubs and Authorities score (Kleinberg 1999) to identify influential actors in the network. Rather than just counting the number of times each actor is identified, this method takes into consideration the “knowledge” that each nominator has of the network. In this way, an influential actor (Authority) is considered as such if more “knowledgeable” actors (Hubs) nominate them. The sum of squared Authority scores is one, so squared Authority scores provide an overall idea of power distribution in the network of actors.

Social Cognitive Mapping can be used to identify groups within a network, and it partly relies on the understanding of social dynamics by a third party. As several members of the network identify existing ties, this tool provides a method of triangulation of information. Additionally, it is useful when not all members of a network participate in the collection of data (Neal 2008). Question 7 in this study aimed at identifying coalitions. The most relevant stage of the analysis (Neal and Neal 2013) was the identification of co-membership groups. The answers were organised in a valued matrix, in which higher numbers indicated more agreement about the existence of the link. An E-I index provides a measure of clustering of subgroups in a network according to specified attributes by counting external links, subtracting internal links and dividing by the total number of links (Hanneman and Riddle 2005). The index value ranges from -1 (all links are internal) to 1 (all links are external). Actors in this study were divided into three categories: those with a specific interest on the extraction of resources (“extractive”), those with a specific interest on conservation (“conservation”), and those with more indirect or ambiguous interests (“other”). As the E-I index only measures the presence or absence of links, the valued matrix was dichotomised, including only values greater than two (three or more people assumed that the actors were part of a coalition). After counting internal and external links, the procedure runs 5,000 permutations, calculating the probability of obtaining the same index by chance.

To understand the social structures across sectors in the Tasmanian MPA information exchange network, I conducted a statistical analysis using ERGMs. For this network, only reciprocated ties from interviewed organisations were included. Standard statistical models assume data independence, while ERGMs account for the inherent dependence of relational data. ERGMs allow the comparison between the observed network and a series of random configurations of similar networks (Robins et al. 2004). In this way, statistical inferences can be made regarding the prevalence or absence of key configurations. To understand relationships across sectors, I used non-directed graphs for actors with attributes (Wang et al. 2009). Attributes followed the same division of actors as for the E-I indexes, namely “extractive”, “conservation” and “other”. First, I ran an estimation of how many of each structure were present in the network. Any configuration that was not present was excluded from models; configurations with low numbers were also explored cautiously and eventually excluded in the model if they were preventing convergence or a reasonable goodness of fit. The best model I ran included structures that indicate “closure” of the network (see Chapter 2 for an explanation) or triangles; structures that indicate brokerage or stars; and links across sectors. These structures are shown in Table 3.3.

Table 3.3 Main social structures analysed in exponential random graph models (ERGMs) in this study

Black circles in structures indicate nodes with specific attributes.

Effects Structure

2-star

Triangle

Cross sector collaboration (extractive)

Cross sector collaboration (conservation)

Cross sector bridging (extractive)

Cross sector bridging (conservation) Cross sector links (extractive) Cross sector links (conservation)

Effects with significant positive estimates indicate that the observed network has more of that configuration than expected by chance, while significant negative values indicate that the configuration occurs less frequently than expected by chance. The presence/absence of different configurations can be interpreted based on the theoretical background (see Chapter 2).