Commentary: Understanding social network analysis

BMJ 2008; 337 doi: (Published 5 December 2008)
Cite this as: BMJ 2008;337:a1957
  1. Peter Sainsbury, director, population health
  1. 1Sydney South West Area Health Service, Locked Bag 7008, Liverpool, NSW 1871, Australia
  1. sainsburyp{at}

    In the linked study (doi:10.1136/bmj.a2338), Fowler and Christakis investigated the new and intriguing hypothesis that people’s happiness is influenced by, among other things, the happiness of their acquaintances, particularly first degree relatives, close friends, neighbours, and coworkers.1 The authors cleverly use the Framingham heart study’s existing database that includes, fortuitously rather than by design, information that can be used for social network analysis. Their results broadly confirm this hypothesis, but many readers will be unfamiliar with social network analysis, confused by the analytical techniques, and unsure about the validity of the findings.

    Humans are unavoidably social beings. Consequently, not only does society exist, but its existence is inevitable, and each person is influenced in many ways by society at large and individuals and groups within it. It follows that to understand the attributes of individuals (for instance their behaviour and health) the research toolkit must include methods that explore the social relationships between people. Social network analysis is one such method.

    Put simply, by asking study participants to list the people they know, and which acquaintances know each other, social network analysis researchers seek to represent visually and analyse quantitatively the web of relationships around and among people. Of course, in reality, it is more complex than this. For instance, researchers may focus on the relationships around each individual or they may aggregate these to construct the more complex web of relationships within a community (for instance a business organisation or a town); or researchers may focus on everyone known to each study participant or, more commonly, on a particular group of their acquaintances (for instance their family or the people they see daily). Depending on their specific aims, researchers carefully phrase their questions to participants (the “name generators”) to identify the types of acquaintances they are interested in.

    Fowler and Christakis’s study has several strengths. Firstly, when the information was collected it was not intended that it would be used to measure happiness, analyse social networks, or explore this hypothesis. Consequently, the original data collection was not biased by the researchers’ desire to confirm this hypothesis or by the participants’ wishes to give socially desirable answers. Secondly, although social network analysis is complex and unfamiliar to many, this research method is commonly used by sociologists, community psychologists, and others. Thirdly, despite the sometimes large and overlapping confidence intervals, the results are internally consistent and robust to sensitivity analyses.

    We should be cautious, however, for several reasons. Firstly, a single community and a single database that was not designed to tackle this hypothesis was studied—perhaps Framingham is unique in some way; perhaps the data collection incorporated an unknown systematic bias that produced these results. Secondly, the findings concerning friends must be viewed cautiously because the name generator used seems unlikely to have encouraged respondents with several close friends to name more than one. From a social network analysis viewpoint it would have been preferable to ask respondents to name all their close friends. This would have generated more complete networks and made it more likely that mutual friends would have been identified. Also, the size of the influence of distant friends (friends of friends’ friends; 5.6%) seems overly large when the influence of a happy friend is only 14%. Thirdly, the measure of happiness is well validated as a measure of “positive affect,” but it will be interesting to see if similar results are produced with different measures of happiness. Fourthly, happiness is not everything; unhappy acquaintances may contribute something other than happiness to our lives.

    In summary, Fowler and Christakis have produced valuable, exciting, and reasonably robust results that will stimulate new and productive lines of enquiry in happiness studies. However, we must not expect all the details of their findings to be confirmed in subsequent work. Don’t drop your unhappy friends yet.


    Cite this as: BMJ 2008;337:a1957


    • Research, doi: 10.1136/bmj.a2338
    • Competing interests: None declared, but PS donated the £100 (€126; $178) fee for writing this article to Amnesty International.

    • Provenance and peer review: Commissioned; not externally peer reviewed.

    This is an open-access article distributed under the terms of the Creative Commons Attribution Non-commercial License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.