Developing networks for community change: exploring the utility of network analysis

Abstract

This article demonstrates how network analysis can be used to develop a better understanding of a community-based network and steps that might be taken to facilitate network development. The focal network in this study was organized by a local non-profit organization as part of their effort to effect community-level change. The activities of this network were guided by three steering committees, 23 members of which provided data regarding numerous aspects of their relationships with a set of 39 organizations. We first provide descriptive information regarding patterns of relationships among network participants at both the committee and whole-network level. We then summarize results of hierarchical linear modeling and analysis of variance analyses that clarify how data generated through network analysis can be incorporated into these traditional analytic procedures to yield additional insights regarding network properties. The article concludes with a discussion of the implications of these findings for the continued development of this community-based network, as well as implications for others interested in using network analysis to facilitate change in their own communities.

Keywords:

• community-based organizations
• network analysis
• health
• social capital

Acknowledgements

This study was conducted under a grant from the Centers for Disease Control and Prevention (02153).

Notes

1. Consistent with ethical guidelines suggested by Borgatti and Molina (2003), results from analysis of the network data were presented to network members who attended a subsequent consortium at CHC.

2. This survey format generated two-mode network data, in contrast to the one-mode data that are more typical in network analysis. These data are two-mode because the set of respondents is different from the set of organizations for which the respondents provided information regarding their linkages. In this case, the result is a 23 x 39 matrix in which each cell provides information regarding the nature of the relationship between one of the 23 respondents (i.e. the organization she/he represents) and one of the 39 organizations on the list. (Actually, seven such matrices are generated from the survey data, one for each of the relationship characteristics assessed.) These two-mode matrices contrast with the square matrices generated by one-mode data, in which respondents provide information about their linkages to each other. Some of the calculations involved in network analysis require the square matrices of a one-mode dataset and thus cannot be done using two-mode data.

3. Since regularity was assessed using a four-point scale, this variable was dichotomized (1–2 vs. 3–4) to specify whether or not there was a regular relationship between a respondent and an organization.

4. Figures 3a and 3b were configured differently simply to demonstrate the interactive versatility of the Netdraw component of the UCINET software, which allows the user to “drag” a node to any spot on the page while maintaining its links to other nodes and thus to shape the network graph however desired. The columnar format reflected in Figure 3a was provided to demonstrate the contrast with the more conventional kind of graph provided in Figure 3b, and with the previous graphs that more explicitly separated respondents from organizations.

5. We wanted to include two dummy variables so that we could contrast non-profit organizations with government and for-profit organizations separately, but the small number of dyads involving a respondent representing a for-profit organization caused problems in the calculations needed to generate that coefficient estimates in hierarchical linear modeling analysis, and thus we decided to include the single variable that allows us to assess whether respondents from non-profit organizations are more or less trusting of other organizations than their counterparts in other sectors.

6. Although the results are not reported here, an analysis including the four relationship types individually, rather than the multiplexity variable, indicates that sharing resources and engaging in policy advocacy together are related to higher trust, while sharing information and joint programming are not.