Page - (000322) - in Knowledge and Networks

320 half of the issues. Thus, SALAR seems to function as an important bridge between local governments. This finding was reinforced in the in-depth interviews. Table 15.1 verifies another conclusion drawn from the in-depth interviews: infor- mal personal connections are the most important channel used to get information about activities in other municipalities. Other important channels include the inter- net homepages of other municipalities, regular meetings and written reports where SALAR describes various activities at the local level. Learning Clusters Are Based on County Structure In an analysis of dyads of municipalities by Lundin et al. (2015) using the same data analyzed in this chapter, it is demonstrated that Swedish municipalities primarily learn from their local neighbors.4 The analysis reveals that geographic proximity greatly increases the probability that a municipality will learn from another munici- pality. For instance, if two municipalities are located in the same county, the pre- dicted probability that a learning link will be established is .054, all else being equal. If the municipalities are not located in the same county, the probability is only .003. Our interviews point in the same direction. As one local civil servant told us: Above all, it is if you find somebody that is successful in a certain area, somebody that has given thought to something. Primarily it is often among neighbors that we look because they are quite similar and work under roughly the same conditions. (2011 interview with civil servant) The importance of geographical proximity suggests that municipal learning net- works in Sweden might be very parochial. If so, it is reasonable to expect that new knowledge, ideas, and best practices would be quite slow to diffuse to local govern- ments in Sweden. But the dyadic analysis presented above does not account for the possibility of learning indirectly through less proximate networks. To explore this possibility, we employ ideas from social network analysis. The aim is to deepen our understanding of the clustering properties of a potential learning network: How are clusters structured? To find out more about this, we started by using the Girvan- Newman method of detecting community structure (Girvan & Newman, 2002). This method finds clusters by iteratively removing edges with high edge betweenness scores until it reaches some specified minimum number of clusters. A betweenness score summarizes the number of times an actor is a bridge between two other actors in the network (Freeman, 1977). The analysis is presented in Table 15.2 and it reinforces findings in Lundin et al. (2015): county is a very strong predictor of cluster membership. There are 21 coun- ties in Sweden and when directed to detect 21 clusters, the community structure 4 The main research question in Lundin et al. (2015) is whether local governments tend to learn from governments that are more successful than others; the empirical findings support this hypoth- esis. However, the importance of proximity, similarity, and power is also examined in the study. C. Ansell et al.