Key Players: Not just the most connected nodes

One of the most common uses of Social Network Analysis in developing country contexts to in identifying central actors within a community who are well-connected and perhaps most able to spread a new idea to others in a community quickly.  These individuals are variously called champions, opinion leaders, or change agents among other labels.  Their connectedness means they are trusted and have a strong influence on the beliefs and behaviors of others in the community.  Many interventions have looked at leveraging these individuals for everything from changing drug prescribing habits of physicians to hybrid seed usage by farmers to family planning methods for women in developing countries.

There are many different methods for finding the most influential individuals, ranging from surveying the entire network and computing them to simpler methods such as asking those in influential positions or friendship nomination.  A recent study looked at basing their intervention targets on mapping out the whole network, computing how many connections each person had (technically, in-degree centrality), and choosing those with the most connections.  This intensive method, one of a few tested, actually didn’t improve outcomes compared to a random selection of individuals.  There could be many reasons for this, but one of the main ones in my opinion is that this is the wrong way to think about targeting.

The authors don’t totally disagree–they recognize many of the pitfalls of this method as well.  Targeting the most connected individuals may tend to target those connected to each other, meaning that parts of the network separated from the most well-connected because of sociodemographic differences or even just spatial distance may be quite far from a targeted individual in terms of the number of ties an intervention must pass through to reach them.

A more promising approach, described by Stephen Borgatti (2006), is to identify “Key Players” in the network.  Rather than just computing a set of people who are the most central to a network, the KPP-Pos algorithm he describes selects a set of individuals such that the distance from individuals in the network to their nearest Key Player is minimized.  Borgatti’s algorithm is based on computing a “distance-weighted reach” of a selected set of nodes, so that how far nodes are from the overall set is considered rather than just looking at the most connected nodes.

A simple example where this might be useful is shown below, where 1 and 2 are the most central individuals based on degree centrality (5 ties each), but if you want to target two individuals, selecting 1 and 7 would yield a better coverage of the network through direct contact with those targeted.  Though this diagram seems a bit contrived, homophily is so strong in networks that similar results are not uncommon in real world data (and two real-world data sets are tested by Borgatti).

KeyPlayers_cropped

It seems a rather straightforward idea, and certainly more complex techniques have been developed since this technique was published, yet the authors of the above study either seem unaware of this technique or think it not so obviously superior to methods used in the study as to merit inclusion.

A question that comes to mind based on these techniques is: Is targeting multiple individuals connected to the same people more effective than targeting just individual Key Players?  Certainly it seems that targeting a group would be more effective than targeting individuals, but is the difference worth the smaller number of groups that could be targeted for a given project cost?  If you know the answer, I’d be happy to hear about it!